From: Lynn Wheeler <lynn@garlic.com>
Subject: Mainframe and Cloud
Date: 06 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025d.html#112 Mainframe and Cloud
... note online/cloud tends to have capacity much greater than avg use
... to meet peek on-demand use which could be order of magnitude
greater. cloud operators had heavily optimized server blade systems
costs (including assembling their own systems for a fraction of brand
name servers) ... and power consumption was increasingly becoming a
major expense. There was then increasing pressure on makers of server
components to optimize power use as well as allowing power use to drop
to zero when idle ... but instant on to meet on-demand requirements.
large cloud operation can have a score (or more) of megadatacenters
around the world, each with half million or more server blades, and
each server blade with ten times rocessing of max. configured
mainframe .... and enormous automation; a megadatacenter with 70-80
staff (upwards of 10,000 or more systems per staff). In the past were
articles about being able to use a credit card to on-demand spin up
for a couple of hrs, a cloud ("cluster") supercomputer (that ranked in
the top 40 in the world)
megadatacenter posts
https://www.garlic.com/~lynn/submisc.html#megadatacenter
GML was invented at the IBM Cambridge Science Center in 1969 (about
the same time that CICS product appeared) .... after decade morphs
into ISO standard SGML and after another decade morphs into HTML at
CERN.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
GML, SGML, HTML posts
https://www.garlic.com/~lynn/submain.html#sgml
CICS/BDAM posts
https://www.garlic.com/~lynn/submain.html#bdam
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Mainframe skills
Date: 06 Oct, 2025
Blog: Facebook
In CP67->VM370 (after decision to add virtual memory to all 370s),
lots of stuff was simplified and/or dropped (including multiprocessor
support) ... adding virtual memory to all 370s
https://www.garlic.com/~lynn/2011d.html#73
Then in 1974 with a VM370R2, I start adding a bunch of stuff back in
for my internal CSC/VM (including kernel re-org for multiprocessor,
but not the actual SMP support). Then with a VM370R3, I add SMP back
in, originally for (online sales&marketing support) US HONE so
they could upgrade all their 168s to 2-CPU 168s (with little slight of
hand getting twice throughput).
Then with the implosion of Future System
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
I get asked to help with a 16-CPU 370, and we con the 3033 processor
engineers into helping in their spare time (a lot more interesting
than remapping 168 logic to 20% faster chips). Everybody thought it
was great until somebody tells the head of POK that it could be
decades before POK favorite son operating system ("MVS") had
(effective) 16-CPU support (MVS docs at the time saying 2-CPU systems
had 1.2-1.5 times throughput of single CPU; POK doesn't ship 16-CPU
system until after turn of century). Then head of POK invites some of
us to never visit POK again and directs 3033 processor engineers heads
down and no distractions.
2nd half of 70s transferring to SJR on west coast, I worked with Jim
Gray and Vera Watson on original SQL/relational, System/R (done on
VM370); we were able to tech transfer ("under the radar" while
corporation was pre-occupied with "EAGLE") to Endicott for
SQL/DS. Then when "EAGLE" implodes, there was request for how fast
could System/R be ported to MVS ... which was eventually released as
DB2, originally for decision-support *only*.
I also got to wander around IBM (and non-IBM) datacenters in silicon
valley, including DISK bldg14 (engineering) and bldg15 (product test)
across the street. They were running pre-scheduled, 7x24, stand-alone
testing and had mentioned recently trying MVS, but it had 15min MTBF
(requiring manual re-ipl) in that environment. I offer to redo I/O
system to make it bullet proof and never fail, allowing any amount of
on-demand testing, greatly improving productivity. Bldg15 then gets
1st engineering 3033 outside POK processor engineering ... and since
testing only took percent or two of CPU, we scrounge up 3830
controller and 3330 string to setup our own private online
service. Then bldg15 also gets engineering 4341 in 1978 and some how
branch hears about it and in Jan1979 I'm con'ed into doing a 4341
benchmark for a national lab that was looking at getting 70 for
compute farm (leading edge of the coming cluster supercomputing
tsunami).
Decade later in 1988, got approval for HA/6000 originally for NYTimes
to port their newspaper system (ATEX) from DEC VAXCluster to
RS/6000. I rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Ingres, Sybase, Informix that have DEC
VAXCluster support in same source base with UNIX). IBM S/88 Product
Administrator was also taking us around to their customers and also
had me write a section for corporate continuous availability strategy
document (it gets pulled when both Rochester/AS400 and POK/mainframe
complain).
Early Jan92 meeting with Oracle CEO, AWD executive Hester tells
Ellison that we would have 16-system clusters by mid92 and 128-system
clusters by ye92. Mid-jan92 convince FSD to bid HA/CMP for
gov. supercomputers. Late-jan92, cluster scale-up is transferred for
announce as IBM Supercomputer (for technical/scientific *only*) and we
were told we couldn't work on clusters with more than four systems (we
leave IBM a few months later).
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to the
industry MIPS reference platform):
• ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
• RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Former executive we had reported to, goes over to head up Somerset/AIM
(Apple, IBM, Motorola), single chip RISC with M88k bus/cache (enabling
clusters of shared memory multiprocessors)
i86 chip makers then do hardware layer that translate i86 instructions
into RISC micro-ops for actual execution (largely negating throughput
difference between RISC and i86); 1999 industry benchmark:
• IBM PowerPC 440: 1,000MIPS
• Pentium3: 2,054MIPS (twice PowerPC 440)
early numbers actual industry benchmarks, later used IBM pubs giving
percent change since previous
z900, 16 processors 2.5BIPS (156MIPS/core), Dec2000
z990, 32 cores, 9BIPS, (281MIPS/core), 2003
z9, 54 cores, 18BIPS (333MIPS/core), July2005
z10, 64 cores, 30BIPS (469MIPS/core), Feb2008
z196, 80 cores, 50BIPS (625MIPS/core), Jul2010
EC12, 101 cores, 75BIPS (743MIPS/core), Aug2012
z13, 140 cores, 100BIPS (710MIPS/core), Jan2015
z14, 170 cores, 150BIPS (862MIPS/core), Aug2017
z15, 190 cores, 190BIPS (1000MIPS/core), Sep2019
z16, 200 cores, 222BIPS (1111MIPS/core), Sep2022
z17, 208 cores, 260BIPS (1250MIPS/core), Jun2025
Also in 1988, the branch office asks if I could help LLNL (national
lab) standardize some serial they were working with, which quickly
becomes fibre-channel standard, "FCS" (not First Customer Ship),
initially 1gbit/sec transfer, full-duplex, aggregate 200mbye/sec
(including some stuff I had done in 1980). Then POK gets some of their
serial stuff released with ES/9000 as ESCON (when it was already
obsolete, initially 10mbytes/sec, later increased to
17mbytes/sec). Then some POK engineers become involved with FCS and
define a heavy-weight protocol that drastically cuts throughput
(eventually released as FICON).
2010, a z196 "Peak I/O" benchmark released, getting 2M IOPS using 104
FICON (20K IOPS/FICON). About the same time a FCS is announced for
E5-2600 server blade claiming over million IOPS (two such FCS having
higher throughput than 104 FICON). Also IBM pubs recommend that SAPs
(system assist processors that actually do I/O) be kept to 70% CPU (or
1.5M IOPS) and no new CKD DASD has been made for decades, all being
simulated on industry standard fixed-block devices. Note: 2010 E5-2600
server blade (16 cores, 31BIPS/core) benchmarked at 500BIPS (ten times
max configured Z196).
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
SMP, tightly-coupled, shared-memory multiprocessor
https://www.garlic.com/~lynn/subtopic.html#smp
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
original SQL/relational posts
https://www.garlic.com/~lynn/submain.html#systemr
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
Fibre Channel Standard ("FCS") &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
DASD, CKD, FBA, multi-track posts
https://www.garlic.com/~lynn/submain.html#dasd
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: VM370 Teddy Bear
Date: 09 Oct, 2025
Blog: Facebook
the SHARE MVS mascot was the "turkey" (because MVS/TSO was so bad) as
opposed to the VM370 mascot as the teddy bear.
https://www.jaymoseley.com/hercules/downloads/pdf/$OSTL33.pdf
pg8:
And let us not forget that the performance of the first releases of
MVS was so bad that the MVS people in the SHARE user group adopted the
Turkey as their symbol.
... vs the SHARE VM370 mascot:
The symbol of the VM group was the Teddy Bear since it was said to be
better, warmer, and more user-friendly than MVS.
trivia1: 1974, CERN did SHARE presentation comparing MVS/TSO and
VM370/CMS ... copies inside IBM were stamped "IBM Confidential -
Restricted", 2nd highest security classification, available on need to
know only.
trivia2: customers not migrating to MVS as planned (I was at the the
initial SHARE when it was played).
http://www.mxg.com/thebuttonman/boney.asptrivia3: after FS imploded:
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
the head of POK managed to convince corporate to kill the VM370
product, shutdown the development group and transfer all the people to
POK for MVS/XA (Endicott eventually managed to save the VM370 product
mission for the mid-range, but had to recreate a development group
from scratch). They weren't planning on telling the people until the
very last minute to minimize the number that might escape. It managed
to leak early and several managed to escape (it was during infancy of
DEC VAX/VMS ... before it even first shipped and joke was that head of
POK was a major contributor to VMS). There was then a hunt for the
leak, fortunately for me, nobody gave up the leaker.
Melinda's history
https://www.leeandmelindavarian.com/Melinda#VMHist
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67l, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Interactive Response
Date: 11 Oct, 2025
Blog: Facebook
I took a 2 credit hr intro to fortran/computers and at end of semester
was hired to rewrite 1401 MPIO in assembler for 360/30 ... univ was
getting 360/67 for tss/360 to replace 709/1401 and got a 360/30
temporarily pending availability of a 360/67 (part of getting some 360
experience). Univ. shutdown datacenter on weekends and I would have it
dedicated (although 48hrs w/o sleep made monday classes hard), I was
given a pile of hardware & software manuals and got to design and
implement my own monitor, device drivers, interrupt handlers, storage
management, error recovery, etc ... and within a few weeks had a 2000
card program.
Then within a yr of taking intro class, 360/67 arrives and i'm hired
fulltime responsible for os/360 (tss/360 never came to production so
ran as 360/65).
CSC comes out to univ for CP67/CMS (precursor to VM370/CMS) install
(3rd after CSC itself and MIT Lincoln Labs) and I mostly get to play
with it during my 48hr weekend dedicated time. I initially work on
pathlengths for running OS/360 in virtual machine. Test stream
ran 322secs on real machine, initially 856secs in virtual
machine (CP67 CPU 534secs), after a couple months I have reduced
CP67 CPU from 534secs to 113secs. I then start rewriting the
dispatcher/scheduler (dynamic adaptive resource manager/default fair
share policy) paging, adding ordered seek queuing (from FIFO) and
mutli-page transfer channel programs (from FIFO and optimized for
transfers/revolution, getting 2301 paging drum from 70-80 4k
transfers/sec to channel transfer peak of 270). Six months after univ
initial install, CSC was giving one week class in LA. I arrive on
Sunday afternoon and asked to teach the class, it turns out that the
people that were going to teach it had resigned the Friday before to
join one of the 60s CP67 commercial online spin-offs.
Before I graduate, I'm hired fulltime into a small group in the Boeing
CFO office to help with form Boeing Computer Services (consolidate all
dataprocessing into an independent business unit ... including
offering services to non-Boeing entities). When I graduate, I leave to
join CSC (instead of staying with CFO).
One of my hobbies after joining IBM was enhance production operating
systems for internal datacenters ... and in the morph of
CP67->VM370 a lot of stuff was dropped and/or simplified and every
few years I would be asked to redo stuff that had been dropped and/or
rewritten (... in part dynamic adaptive default policy calculated
dispatching order based on resource utilization over the last several
minutes compared to target resource utilization established by their
priority and number of users).
Late 80s, the OS2 team was told to contact VM370 group (because VM370
dispatching was much better than OS2) ... it was passed between the
various groups before being forwarded to me.
Example I didn't have much control over was late 70s, IBM San Jose
Research got a MVS 168 and a VM370 158 replacing MVT 195. My internal
VM370s were getting 90th percentile .11sec interactive system response
(with 3272/3277 hardware response of .086sec resulted in .196sec seen
by users ... better than the .25sec requirement mentioned in various
studies). All the SJR 3830 controllers and 3330 strings were
dual-channel connection to both systems but strong rules that no MVS
3330s can be mounted on VM370 strings. One morning operators mounted a
MVS 3330 on a VM370 string and within minutes they were getting irate
calls from all over the bldg complaining about response. The issue was
MVS has a OS/360 heritage of multi-track search for PDS directory
searches ... a MVS multi-cylinder PDS directory search can have
multiple full multi-track cylinder searches that lockout the (vm370)
controller for the duration (60revs/sec, 19tracks/search, .317secs
lockout per multi-track search I/O). Demand to move the pack was
answered with they would get around to it on 2nd shift.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
dynamic adaptive resource dispatch/scheduling posts
https://www.garlic.com/~lynn/subtopic.html#fairshare
commercial virtual machine service posts
https://www.garlic.com/~lynn/submain.html#timeshare
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
DASD, CKD, FBA, multi-track search posts
https://www.garlic.com/~lynn/submain.html#dasd
some posts mentioning .11sec system response and .086sec
3272/3277 hardware response for .196sec
https://www.garlic.com/~lynn/2025c.html#6 Interactive Response
https://www.garlic.com/~lynn/2025.html#69 old pharts, Multics vs Unix
https://www.garlic.com/~lynn/2024d.html#13 MVS/ISPF Editor
https://www.garlic.com/~lynn/2022c.html#68 IBM Mainframe market was Re: Approximate reciprocals
https://www.garlic.com/~lynn/2021j.html#74 IBM 3278
https://www.garlic.com/~lynn/2021i.html#69 IBM MYTE
https://www.garlic.com/~lynn/2016e.html#51 How the internet was invented
https://www.garlic.com/~lynn/2016d.html#104 Is it a lost cause?
https://www.garlic.com/~lynn/2016c.html#8 You count as an old-timer if (was Re: Origin of the phrase "XYZZY")
https://www.garlic.com/~lynn/2015g.html#58 [Poll] Computing favorities
https://www.garlic.com/~lynn/2014g.html#26 Fifty Years of BASIC, the Programming Language That Made Computers Personal
https://www.garlic.com/~lynn/2013g.html#14 Tech Time Warp of the Week: The 50-Pound Portable PC, 1977
https://www.garlic.com/~lynn/2012p.html#1 3270 response & channel throughput
https://www.garlic.com/~lynn/2012.html#15 Who originated the phrase "user-friendly"?
https://www.garlic.com/~lynn/2012.html#13 From Who originated the phrase "user-friendly"?
https://www.garlic.com/~lynn/2011p.html#61 Migration off mainframe
https://www.garlic.com/~lynn/2011g.html#43 My first mainframe experience
https://www.garlic.com/~lynn/2010b.html#31 Happy DEC-10 Day
https://www.garlic.com/~lynn/2009q.html#72 Now is time for banks to replace core system according to Accenture
https://www.garlic.com/~lynn/2009q.html#53 The 50th Anniversary of the Legendary IBM 1401
https://www.garlic.com/~lynn/2009e.html#19 Architectural Diversity
https://www.garlic.com/~lynn/2006s.html#42 Ranking of non-IBM mainframe builders?
https://www.garlic.com/~lynn/2005r.html#15 Intel strikes back with a parallel x86 design
https://www.garlic.com/~lynn/2005r.html#12 Intel strikes back with a parallel x86 design
https://www.garlic.com/~lynn/2001m.html#19 3270 protocol
--
virtualization experience starting Jan1968, online at home since Mar1970
Interesting timing issues on 1970s-vintage IBM mainframes
From: Lynn Wheeler <lynn@garlic.com>
Subject: Re: Interesting timing issues on 1970s-vintage IBM mainframes
Newsgroups: alt.folklore.computers
Date: Sun, 12 Oct 2025 08:24:39 -1000
James Dow Allen <user4353@newsgrouper.org.invalid> writes:
Interest was minimal in the (startling?) fact that the main 145 oscillator
is immediately passed through an XOR gate. But I'll persist and mention
interesting(?) facts about the clocks on the Big IBM Multiprocessors.
The 370/168 was, arguably, the Top Of The Line among IBM mainframes
in the mid-1970s. Sure, there was a 370 Model 195 but it was almost just
a gimmick: Salesmen might say "You're complaining about the $3 Million
price-tag on our wonderful Model 168?
Just be thankful I'm not trying to sell you a Model 195!"
After joining IBM, the 195 group talk me into helping with
hyperthreading the 195. 195 had out-of-order, but conditional branches
drained the pipeline ... so most codes only ran at half the rated
speed. hyperthreading, simulating 2CPU multiprocessor possibly would
keep the hardware fully busy ... hyperthreading patent mentioned in
this about the death of ACS/360 (Amdahl had won the battle to make
ACS, 360 compatible, the ACS/360 was killed, folklore was executives
felt it would advance the state of art to fast and company would loose
control of the market).
https://people.computing.clemson.edu/~mark/acs_end.html
modulo MVT (VS2/SVS & VS2/MVS) documentation (heavy-weight
multiprocessor overhead) SMP, only had 2-CPU throughtput 1.2-1.5 times
single processor throughput.
early last decade, I was asked to track down the decision to add
virtual memory to all 370s (pieces, originally posted here and in
ibm-main NGs) ... adding virtual memory to all 370s
https://www.garlic.com/~lynn/2011d.html#73
Basically MVT storage management was so bad that region sizes had to
be specified four times larger than used ... as result typical 1mbyte
370/165 only ran four concurrent regions, insufficient to keep system
busy and justified. Going to single 16mbyte virtual address space
(i.e. VS2/SVS ... sort of like running MVT in a CP67
16mbyte virtual machine) allowed concurrent regions to be
increased by factor of four (modulo caped at 15 because 4bit storage
protect keys) with little or no paging.
It was deemed that it wasn't worth the effort to add virtual memory to
370/195 and all new work was killed.
Then there was the FS effort, going to completely replace 370 and
internal politics was killiing off 370 efforts, claims that lack of new
370s during FS gave the clone 370 makers their market foothold).
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
Note 370/165 avg 2.1 machine cycles per 370 instruction. for 168 they
significantly increase main memory size & speed and microcode was
optimized resulting in avg of 1.6 machine cycles per instruction.
Then for 168-3, they doubled the size of processor cache, increasing
rated MIPS from 2.5MIPS to 3.0MIPS.
With the implosion of FS there was mad rush to get stuff back into the
370 product pipelines, kicking off the quick&dirty 3033 and 3081
efforts. The 3033 started off remapping 168 logic to 20% faster chips
and then optimized the microcode getting it down to avg of one machine
cycle per 370 instruction.
I was also talked into helping with a 16-CPU SMP/multiprocessor effort
and we con the 3033 processor engineers into helping (a lot more
interesting than remapping 168 logic). Everybody thought it was great
until somebody reminds the head of POK that POK's favorite son
operating system ("VS2/MVS") 2CPU multiprocessor overhead only getting
1.2-1.5 times throughput of non-multiprocessor (and overhead
increasing significantly as #CPUs increased ... POK doesn't ship a
16-CPU machine until after the turn of century). Then head of POK
invites some of us to never visit POK again and directs the 3033
processor engineers, heads down and no distractions.
trivia: when I graduate and join IBM Cambridge Science Center, one of
my hobbies was enhanced production operating systems and one of my
first (and long time) customers was the online sales&marketing
HONE systems. With the decision to add virtual memory to all 370s,
there was also decision to form development group to do VM370. In the
morph of CP67->VM370, lots of stuff was simplified and/or dropped
(including multiprocessor support). 1974, I start adding stuff back
into a VM370R2-base for my interal CSC/VM (including kernel-reorg for
SMP, but not the actual SMP support). Then for VM370R3-base CSC/VM, I
add multiprocessor support back in, originally for HONE so they could
upgrade their 168s to 2-CPU systems (with some slight-of-hand and
cache affinity, was getting twice throughput of single processor).
other trivia: US HONE had consolidated all their datacenters in
silicon valley, when FACEBOOK first moved into silicon valley, it was
into new bldg built next door to the former consolidated US HONE
datacenter.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
SMP, tightly-coupled, shared-memory multiprocessor
https://www.garlic.com/~lynn/subtopic.html#smp
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
--
virtualization experience starting Jan1968, online at home since Mar1970
Interesting timing issues on 1970s-vintage IBM mainframes
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM CP67 Multi-leavel Update
Date: 12 Oct, 2025
Blog: Facebook
Some of the MIT CTSS/7094 people went to the 5th flr to do
MULTICS. Others went to the IBM Science Center on the 4th flr and did
virtual machines (1st modified 360/40 w/virtual memory and did
CP40/CMS, morphs into CP67/CMS when 360/67 standard with virtual
memory becomes available), science center wide-area network (that
grows into corporate internal network, larger than arpanet/internet
from science-center beginning until sometime mid/late 80s; technology
also used for the corporate sponsored univ BITNET), invented GML 1969
(precursor to SGML and HTML), lots of performance tools, etc.
I took a 2 credit hr intro to fortran/computers and at end of semester
was hired to rewrite 1401 MPIO in assembler for 360/30 ... univ was
getting 360/67 for tss/360 to replace 709/1401 and got a 360/30
temporarily pending availability of a 360/67 (part of getting some 360
experience). Univ. shutdown datacenter on weekends and I would have it
dedicated (although 48hrs w/o sleep made monday classes hard), I was
given a pile of hardware & software manuals and got to design and
implement my own monitor, device drivers, interrupt handlers, storage
management, error recovery, etc ... and within a few weeks had a 2000
card program. 360/67 arrived within a year of taking intro class and I
was hired fulltime responsible for OS/360 (tss/360 never came to
production, so ran as 360/65). Student Fortran jobs ran under second
on 709. Initially MFTR9.5 ran well over minute. I install HASP cutting
time in half. I then start redoing MFTR11 STAGE2 SYSGEN to carefully
place datasets and PDS members to optimize arm seek and multi-track
seach cutting another 2/3rds to 12.9secs. Student Fortran never got
better than 709 until I install UofWaterloo WATFOR (on 360/65 ran at
20,000 cards/min or 333q cards/sec, student Fortran jobs typically
30-60 cards).
CSC came out to univ for CP67/CMS (precursor to VM370/CMS) install
(3rd after CSC itself and MIT Lincoln Labs) and I mostly get to play
with it during my 48hr weekend dedicated time. I initially work on
pathlengths for running OS/360 in virtual machine. Test stream ran
322secs on real machine, initially 856secs in virtual machine (CP67
CPU 534secs), after a couple months I have reduced that CP67 CPU from
534secs to 113secs. I then start rewriting the dispatcher, (dynamic
adaptive resource manager/default fair share policy) scheduler,
paging, adding ordered seek queuing (from FIFO) and mutli-page
transfer channel programs (from FIFO and optimized for
transfers/revolution, getting 2301 paging drum from 70-80 4k
transfers/sec to channel transfer peak of 270). Six months after univ
initial install, CSC was giving one week class in LA. I arrive on
Sunday afternoon and asked to teach the class, it turns out that the
people that were going to teach it had resigned the Friday before to
join one of the 60s CP67 commercial online spin-offs.
Initially CP67 source was deliivered on OS/360, source modified,
assembled, txt decks collected, marked with stripe & name across top,
all fit in card tray, BPS loader placed in front and IPLed ... would
write memory image to disk for system IPL. A couple months later, new
release now resident on CMS ... modifications in CMS "UPDATE" files,
exec that applied update and generated temp file that was assembled. A
system generation exec, "punched" txt decks spooled to virtual reader
that was then IPLed.
After graduating and joining CSC, one of my hobbies was enhanced
production operating systems ("CP67L") for internal datacenters
(inluding online sales&marketing support HONE systems, was one of the
first, and long time customer). With the decision to add virtual
memory to all 370s, there was also decision to do CP67->VM370 and some
of the CSC people went to the 3rd flr, taking over the IBM Boston
Programming Center for the VM370 group. CSC developed set of CP67
updates that provided (simulated) VM370 virtual machines
("CP67H"). Then there were a set of CP67 updates that ran on 370
virtual memory architecture ("CP67I"). At CSC, because there were
profs, staff, and students from Boston area institutions using the CSC
systems, CSC would run "CP67H" in a 360/67 virtual machine (to
minimize unannounced 370 virtual memory leaking).
CP67L ran on real 360/67
... CP67H ran in a CP67L 360/67 virtual machine
...... CP67I ran in a CP67H 370 virtual machine
CP67I was in general use, a year before the 1st engineering 370 (with
virtual memory) was operation ... in fact, IPL'ing CP67I on the real
machine was test case.
As part of CP67L, CP67H, CP67I effort, the CMS Update execs were
improved to support multi-level update operation (later multi-level
update support was added to various editors). Three engineers come out
from San Jose and add 2305 & 3330 support to CP67I, creating CP67SJ
which was widely use on internal machines, even after VM370 was
available.
Mid-80s, Melinda
https://www.leeandmelindavarian.com/Melinda#VMHist
asked if I could send her the original exec multi-level update
implementation. I had large archive dating back to the 60s on triple
redundant tapes in the IBM Almaden Research tape library. It was
fortunate since within a few weeks, Almaden had an operation problem
mounting random tapes as scratch and I lost nearly dozen tapes,
including triple redundant tape archive.
In the morph of CP67->VM370, a lot of stuff was simplified and/or
dropped (including multiprocessor support). 1974, I start adding a lot
of stuff back into VM370R2-base for my internal CSC/VM (including
kernel re-org for SMP, but not the actual SMP support). Then with
VM370R3-base, I add multiprocessor support into CSC/VM, initially for
HONE so they could upgrade all their 168 systems to 2-CPU (getting
twice throughput of 1-CPU systems). HONE trivia: All the US HONE
datacenters had been consolidated in Palo Alto ... when FACEBOOK 1st
moved into silicon valley, it was into a new bldg built next door to
the former consolidate US HONE datacenter.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
SMP, tightly-coupled, shared-memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
some CP67L, CP67H, CP67I, CP67SJ, CSC/VM posts
https://www.garlic.com/~lynn/2025d.html#91 IBM VM370 And Pascal
https://www.garlic.com/~lynn/2025.html#122 Clone 370 System Makers
https://www.garlic.com/~lynn/2025.html#121 Clone 370 System Makers
https://www.garlic.com/~lynn/2025.html#120 Microcode and Virtual Machine
https://www.garlic.com/~lynn/2024g.html#108 IBM 370 Virtual Storage
https://www.garlic.com/~lynn/2024g.html#73 Early Email
https://www.garlic.com/~lynn/2024f.html#112 IBM Email and PROFS
https://www.garlic.com/~lynn/2024f.html#80 CP67 And Source Update
https://www.garlic.com/~lynn/2024f.html#29 IBM 370 Virtual memory
https://www.garlic.com/~lynn/2024d.html#68 ARPANET & IBM Internal Network
https://www.garlic.com/~lynn/2023e.html#70 The IBM System/360 Revolution
https://www.garlic.com/~lynn/2023d.html#98 IBM DASD, Virtual Memory
https://www.garlic.com/~lynn/2022h.html#22 370 virtual memory
https://www.garlic.com/~lynn/2022.html#55 Precursor to current virtual machines and containers
https://www.garlic.com/~lynn/2022.html#12 Programming Skills
https://www.garlic.com/~lynn/2021g.html#34 IBM Fan-fold cards
https://www.garlic.com/~lynn/2014d.html#57 Difference between MVS and z / OS systems
https://www.garlic.com/~lynn/2006.html#38 Is VIO mandatory?
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM DASD, CKD and FBA
Date: 13 Oct, 2025
Blog: Facebook
When I offered the MVS group fully tested FBA support ... they said I
needed $26M incremental new revenue (some $200M in sales) to cover
cost of education and documentation. However, since IBM at the time,
was selling every disk it made, FBA support would just move some CKD
sales to FBA ... also I wasn't allowed to use life-time savings as
part of the business case.
All disk technology was actually moving to FBA ... can be seen in 3380
formulas for records/track calculations, having to round record sizes
up to multiple of fixed cell size. Now no real CKD have been made for
decades, all being simulated on industry standard fixed-block
devices. A big part of FBA was error correcting technology performance
... part of recent FBA technology moving from 512byte blocks to 4k
blocks.
trivia: 80s, large corporations were ordering hundreds of vm/4341s at
a time for deploying out in departmental areas (sort of the leading
edge of the coming distributed departmental computing tsunami). Inside
IBM, conference rooms were becoming scarce, being converted into
departmental vm/4341 rooms. MVS looked at the big explosion in sales
and wanted a piece of the market. However the only new non-datacenter
disks were FBA/3370 ... eventually get CKD emulation as 3375. However
didn't do MVS much good, distributed departmental dataprocessing was
looking at scores of systems per support person ... while MVS was
scores of support people per system.
Note: ECKD was originally channel commands for Calypso, 3880 speed
matching buffer allowing 3mbyte/sec 3380 to be used with existing
1.5mbyte/sec channels ... it went through significant teething
problems ... lots and lots of sev1.
Other trivia: I had transferred from CSC out to SJR on west coast and
got to wander around IBM (and non-IBM) datacenters in silicon valley,
including disk bldg14 (engineering) and bldg15 (product test) across
the street. They were running pre-scheduled, 7x24, stand-alone
mainframe testing and mentioned that they had recently tried MVS, but
it had 15min MTBF (requiring manual re-ipl) in that environment. I
offered to rewrite I/O supervisor to make it bullet proof and never
fail allowing any amount of on-demand, concurrent testing ... greatly
improving productivity. I write an internal IBM paper on the I/O
integrity work and mention the MVS 15min MTBF ... bringing down the
wrath of the MVS organization on my head. Later, a few months before
3880/3380 FCS, FE (field engineering) had test of 57 simulated errors
that were likely to occur and MVS was failing in all 57 cases
(requiring manual re-ipl) and in 2/3rds of the cases no indication of
what caused the failure.
DASD, CKD, FBA, and multi-track search posts
https://www.garlic.com/~lynn/submain.html#dasd
getting to play disk engineer in bldgs 14&15 posts
https://www.garlic.com/~lynn/subtopic.html#disk
a few posts mentioning calypso, eckd, mtbf
https://www.garlic.com/~lynn/2025b.html#12 IBM 3880, 3380, Data-streaming
https://www.garlic.com/~lynn/2024g.html#3 IBM CKD DASD
https://www.garlic.com/~lynn/2023c.html#103 IBM Term "DASD"
https://www.garlic.com/~lynn/2015f.html#89 Formal definition of Speed Matching Buffer
a few posts mentioning FBA fixed-block 512 4k
https://www.garlic.com/~lynn/2021i.html#29 OoO S/360 descendants
https://www.garlic.com/~lynn/2019c.html#70 2301, 2303, 2305-1, 2305-2, paging, etc
https://www.garlic.com/~lynn/2017f.html#39 MVS vs HASP vs JES (was 2821)
https://www.garlic.com/~lynn/2014g.html#84 real vs. emulated CKD
https://www.garlic.com/~lynn/2012j.html#12 Can anybody give me a clear idea about Cloud Computing in MAINFRAME ?
https://www.garlic.com/~lynn/2010d.html#9 PDS vs. PDSE
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM DASD, CKD and FBA
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM DASD, CKD and FBA
Date: 13 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#14 IBM DASD, CKD and FBA
semi-related, 1988 IBM branch asks if I could help LLNL (national lab)
standardize some serial stuff they were working with ... which quickly
becomes fibre-channel standard ("FCS", including some stuff I had done
in 1980), initially 1gbit/sec transfer, full-duplex, 200mbyte/sec.
Then POK finally gets their serial stuff released (when it is already
obsolete), initially 10mbyte/sec, later improved to 17mbyte/sec.
Some POK engineers then become involved with FCS and define a
heavy-weight protocol that radically reduces throughput, eventually
released as FICON
Latest public benchmark I've seen is 2010 z196 "Peak I/O", getting 2M
IOPS using 104 FICON (20K IOPS/FICON). About the same time a FCS is
announced for E5-2600 server blades claiming over million IOPS (two
such FCS having higher throughput than 104 FICON). Also IBM pubs
recommend that SAPs (system assist processors that actually do I/O) be
kept to 70% CPU (or 1.5M IOPS).
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM DASD, CKD and FBA
Date: 13 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#14 IBM DASD, CKD and FBA
https://www.garlic.com/~lynn/2025e.html#15 IBM DASD, CKD and FBA
co-worker at CSC was responsible for the CP67-based wide-area network
for the science centers
https://en.wikipedia.org/wiki/Edson_Hendricks
In June 1975, MIT Professor Jerry Saltzer accompanied Hendricks to
DARPA, where Hendricks described his innovations to the principal
scientist, Dr. Vinton Cerf. Later that year in September 15-19 of 75,
Cerf and Hendricks were the only two delegates from the United States,
to attend a workshop on Data Communications at the International
Institute for Applied Systems Analysis, 2361 Laxenburg Austria where
again, Hendricks spoke publicly about his innovative design which
paved the way to the Internet as we know it today.
... snip ...
newspaper article about some of Edson's IBM TCP/IP battles:
https://web.archive.org/web/20000124004147/http://www1.sjmercury.com/svtech/columns/gillmor/docs/dg092499.htm
Also from wayback machine, some additional (IBM missed) references
from Ed's website
https://web.archive.org/web/20000115185349/http://www.edh.net/bungle.htm
... CP67-baseed wide-area network reference by one of the inventors of
GML at the science center
https://web.archive.org/web/20230402212558/http://www.sgmlsource.com/history/jasis.htm
Actually, the law office application was the original motivation for
the project, something I was allowed to do part-time because of my
knowledge of the user requirements. My real job was to encourage the
staffs of the various scientific centers to make use of the
CP-67-based Wide Area Network that was centered in Cambridge.
... snip ...
... morphs into the IBM internal network, larger than arpanet/internet
from just about the beginning until sometime mid/late 80s when forced
to convert to SNA/VTAM. Technology also used for the corporate
sponsored univ BITNET (& EARN in Europe) ... and the VM/4341s
distributed departmental systems ... until change over to large PC and
workstation servers (again in part because of IBM pressure to move to
SNA/VTAM).
then late 80s, senior disk engineer gets a talk scheduled at annual,
internal, world-wide communication group conference, supposedly on
3174 performance. However, the opening was that the communication
group was going to be responsible for the demise of the disk
division. The disk division was seeing drop in disk sales with data
fleeing mainframe datacenters to more distributed computing friendly
platforms. The disk division had come up with a number of solutions,
but they were constantly being vetoed by the communication group (with
their corporate ownership of everything that crossed the datacenter
walls) trying to protect their dumb terminal paradigm. Senior disk
software executive partial countermeasure was investing in
distributed computing startups that would use IBM disks (he would
periodically ask us to drop in on his investments to see if we could
offer any assistance).
The communication group stranglehold on mainframe datacenters wasn't
just disk and a couple years later, IBM has one of the largest losses
in the history of US companies and was being reorganized into the 13
"baby blues" in preparation for breaking up the company ("baby blues"
take-off on the "baby bell" breakup decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
20yrs earlier, Learson tried (and failed) to block bureaucrats,
careerists, and MBAs from destroying Watson culture/legacy, pg160-163,
30yrs of management briefings 1958-1988
https://bitsavers.org/pdf/ibm/generalInfo/IBM_Thirty_Years_of_Mangement_Briefings_1958-1988.pdf
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
GML, SGML, HTML, etc posts
https://www.garlic.com/~lynn/submain.html#sgml
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
BITNET/EARN posts
https://www.garlic.com/~lynn/subnetwork.html#bitnet
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Mainframe TCP/IP and RFC1044
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Mainframe TCP/IP and RFC1044
Date: 13 Oct, 2025
Blog: Facebook
Communication group was fighting to block release of mainframe TCP/IP
support .... when they lost, they changed their tactic and said that
since they had corporate ownership of everything that crosse the
datacenter walls, it had to be released through them, what shipped had
aggregate 44kbytes/sec using nearly whole 3090 processor. It was
eventually ported to MVS by simulating some VM370 diagnose
instructions.
I then added support for RFC1044 and in some tuning tests at Cray
Research between Cray and 4341, got sustained 4341 channel throughput
using only modest amount of 4341 processor (something like 500 times
improvement in bytes moved per instruction executed). Part of the
difference was 8232 was configured as bridge .... while RFC1044
supported mainframe channel attached TCP/IP router (for about same
price as 8232, channel attached router could support dozen ethernet
interfaces, T1&T3, FDDI, and other).
Also con'ed one of the high-speed router vendors into adding support
for RS6000 "SLA" (serial link adapter ... sort of enhanced ESCON,
faster, full-duplex, capable of aggregate 440mbits/sec, they had to
buy the "SLA" chips from IBM) and planning for fiber-channel standard
("FCS"). Part of the original issues was RS6000 SLAs would only talk
to other RS6000s.
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
801/risc, iliad, romp, rios, pc/rt, rs/6000, power, power/pc posts
https://www.garlic.com/~lynn/subtopic.html#801
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Mainframe TCP/IP and RFC1044
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Mainframe TCP/IP and RFC1044
Date: 14 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#18 IBM Mainframe TCP/IP and RFC1044
note early 90s, CSG hired silicon valley contractor to implement
TCP/IP directly in VTAM. What he initially demo'ed had TCP much faster
than LU6.2. He was then told that everybody knows that a "proper"
TCP/IP implementation is much slower than LU6.2 and they would only be
paying for a "proper" implementation.
Senior disk engineer gets talk scheduled at annual, internal,
world-wide communication group conference and opens with the statement
that communication group was going to be responsible for the demise of
the disk division ... datacenter stranglehold wasn't just disk
division and couple years later IBM has one of the largest losses in
the history of US companies ... and was being re-orged into the 13
"baby blues" in preperation for breaking up the company ... see lot
more at a post comment in "public" mainframe group:
https://www.garlic.com/~lynn/2025e.html#17 IBM DASD, CKD and FBA
... also OSI: The Internet That Wasn't. How TCP/IP eclipsed the Open
Systems Interconnection standards to become the global protocol for
computer networking
https://spectrum.ieee.org/osi-the-internet-that-wasnt
Meanwhile, IBM representatives, led by the company's capable director
of standards, Joseph De Blasi, masterfully steered the discussion,
keeping OSI's development in line with IBM's own business
interests. Computer scientist John Day, who designed protocols for the
ARPANET, was a key member of the U.S. delegation. In his 2008 book
Patterns in Network Architecture(Prentice Hall), Day recalled that IBM
representatives expertly intervened in disputes between delegates
"fighting over who would get a piece of the pie.... IBM played them
like a violin. It was truly magical to watch."
... snip ...
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM HASP & JES2 Networking
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM HASP & JES2 Networking
Date: 14 Oct, 2025
Blog: Facebook
recent comment in another post (in this group)
https://www.garlic.com/~lynn/2025e.html#17 IBM DASD, CKD and FBA
co-worker at the science center was responsible for science center
wide-area network that morphs into the corporate internal network and
technology also used for the corporate sponsored univ BITNET (&
EARN in Europe).
When went to try and announce VNET/RSCS for customers it was blocked
by head of POK, this was after the FS implosion
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
and head of POK was in the process of convincing corporate to kill the
VM370 product, shutdown the development group and transfer all the
people to POK for MVS/XA ... and so he was never going to agree to
announcing new VM370-related products (Endicott eventually manages to
acquire the VM370 product mission, but had to recreate a development
group from scratch). This was also after 23jun1969 unbundling announce
and charging for software (with requirement that software revenue had
to cover originally development, and ongoing support and maint) JES2
network code (from HASP that originally carried "TUCC" in cols 68-71)
couldn't meet the revenue requirement ... standard process was
forecast the sales at low, medium, & high ($300/$600/$1200 per
month) price ... and there was no NJE price at which revenue met the
requirement. They then came up with the idea of announcing JES2
Networking & RSCS/VNET as a combined product (merged expenses and
revenue) ... where the RSCS/VNET revenue (which had acceptable
forecast at $30/month) was able to cover JES2 networking.
RSCS/VNET was clean layered design and a NJE emulation driver was easy
to do to connect JES2 into the RSCS/VNET network. However, JES2 still
had to be restricted to boundary nodes: 1) the original HASP used
spare entries in the 255 psuedo device table, usually 160-180 and the
internal network was approaching 768 (and NJE would trash traffice
where the origin or destination node weren't defined), 2) it also
somewhat intermixed network & job control info in header fields,
traffic between two JES2 systems at different release levels had a
habit of crashing the destination MVS, as a result a body of RSCS/VNET
emulated NJE drivers grew up that could recognize header versions and
if necessary reorganize the fileds to keep a destination MVS system
from crashing (there is the infamous case of new SAN Jose JES2 system
crashing a Hursley MVS system ... blamed on the Hursley RSCS/VNET
because they hadn't updated the Hursley RSCS/VNET NJE driver with the
most recent updates to account for San Jose new JES2).
There is also the story of trying to set-up offshift use between San
Jose STL (west coast) and Hursley (England) with a double-hop
satellite circuit. Initially it was brought up between two RSCS/VNET
systems and everything worked fine. Then a STL executive (steeped in
MVS), insisted the circuit be brought up between two JES2 systems
... and nothing worked. They then dropped back to the two RSCS/VNET
systems and everything worked. The executive then commented that
RSCS/VNET must be too dumb to realize it wasn't really working
(despite valid traffic flowing fine in both directions).
other trivia: at the univ, I had taken two credit hr intro to
Fortran/Computers and at end of semester was hired to rewrite 1401
MPIO for 360/30. The univ was getting 360/67 for tss/360 and got
360/30 temporarily replacing 1401 until 360/67 was available. Within a
year of taking intro class, 360/67 arrives and I'm hired fulltime
responsible for OS/360 (tss/360 never came to production). Student
Fortran jobs had run under second on 709 and well over a minute
w/MFTr9.5. I install HASP and cuts the time in half. I then redo
STAGE2 SYSGEN MFTr11 to carefully place datasets and order PDS members
to optimize arm seek and multi-track search, cutting another 2/3rds to
12.9secs (student fortran never got better than 709 until I install
UofWaterloo WATFOR). Later for MVT18 HASP, I strip out 2780 support
(to reduce real storage requirements) and add-in 2741&TTY terminal
support and editor that implemented the CMS EDITOR syntax (code
rewritten from scratch since the environments were so different) for
CRJE support.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
HASP/JES2, ASP/JES3, NJE/NJI posts
https://www.garlic.com/~lynn/submain.html#hasp
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
23jun1969 unbundling announce posts
https://www.garlic.com/~lynn/submain.html#unbundle
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Token-Ring
Date: 15 Oct, 2025
Blog: Facebook
AWD had done their own cards for PC/RT (16bit PC/AT bus) ... including
4mbit Token-Ring card. For the RS/6000 (w/microchannel, 32bit bus),
AWD was told they couldn't do their own cards, but had to use the
(communication group heavily performance kneecapped) PS2 cards. It
turns out the PS2 microchannel 16mbit Token-Ring card had lower
throughput than the PC/RT 4mbit Token-Ring card (i.e. joke that PC/RT
server with 4mbit T/R would have higher throughput than RS/6000 server
with 16mbit T/R).
New Almaden Research bldg had been heavily provisioned with CAT wiring
assuming 16mbit token-ring, but found 10mbit ethernet (over CAT
wiring) LAN had lower latency and higher aggregate throughput than
16mbit token-ring LAN. They also found $69 10mbit ethernet cards had
higher throughput than the PS2 $800 16mbit token-ring cards. We were
out giving customer executive presentations on TCP/IP, 10mbit
ethernet, 3-tier architecture, high-performance routers, distributed
computing presentations (including comparisons with standard IBM
offerings) and taking misinformation barbs in the back by the SNA,
SAA, & token-ring forces. The Dallas E&S center published something
purported to be 16mbit T/R compared to Ethernet ... but only (remotely
valid) explanation I could give was that they compared to early 3mbit
ethernet prototype predating listen-before-transmit (CSMA/CD) part of
Ethernet protocol standard.
About the same time, senior disk engineer gets talk scheduled at
annual, internal, world-wide communication group conference,
supposedly on 3174 performance. However, his opening was that the
communication group was going to be responsible for the demise of the
disk division. The disk division was seeing drop in disk sales with
data fleeing mainframe datacenters to more distributed computing
friendly platforms. The disk division had come up with a number of
solutions, but they were constantly being vetoed by the communication
group (with their corporate ownership of everything that crossed the
datacenter walls) trying to protect their dumb terminal paradigm.
Senior disk software executive partial countermeasure was investing in
distributed computing startups that would use IBM disks (he would
periodically ask us to drop in on his investments to see if we could
offer any assistance). The communication group stranglehold on
mainframe datacenters wasn't just disk and a couple years later, IBM
has one of the largest losses in the history of US companies and was
being reorganized into the 13 "baby blues" in preparation for breaking
up the company ("baby blues" take-off on the "baby bell" breakup
decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
801/risc, iliad, romp, rios, pc/rt, rs/6000, power, power/pc posts
https://www.garlic.com/~lynn/subtopic.html#801
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Token-Ring
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Token-Ring
Date: 15 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#21 IBM Token-Ring
also OSI: The Internet That Wasn't. How TCP/IP eclipsed the Open
Systems Interconnection standards to become the global protocol for
computer networking
https://spectrum.ieee.org/osi-the-internet-that-wasnt
Meanwhile, IBM representatives, led by the company's capable director
of standards, Joseph De Blasi, masterfully steered the discussion,
keeping OSI's development in line with IBM's own business
interests. Computer scientist John Day, who designed protocols for the
ARPANET, was a key member of the U.S. delegation. In his 2008 book
Patterns in Network Architecture(Prentice Hall), Day recalled that IBM
representatives expertly intervened in disputes between delegates
"fighting over who would get a piece of the pie.... IBM played them
like a violin. It was truly magical to watch."
... snip ...
2nd half 80s, I served on (Chessin's) XTP TAB and there were several
gov agencies participating in XTP ... so took XTP (as "HSP") to (ISO
chartered) ANSI X3S3.3 standards group (transport and
network). Eventually was told they were only allowed to do standards
for things that conform to OSI model; XTP didn't because 1) supported
internetworking (non-existed layer between 3&4), 2) bypassed layer 3/4
interface going directly to LAN MAC, and 3) supported LAN MAC,
non-existed layer somewhere in the middle of layer 3. Joke at the time
was Internet/IETF required two interoperable implementations before
standards progression while ISO didn't even require a standard to be
implementable.
XTP/HSP posts
https://www.garlic.com/~lynn/subnetwork.html#xtphsp
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Token-Ring
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Token-Ring
Date: 15 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#21 IBM Token-Ring
https://www.garlic.com/~lynn/2025e.html#22 IBM Token-Ring
trivia: 1988, Nick Donofrio approved HA/6000, originally for NYTimes
to move their newspaper system (ATEX) off DEC VAXCluster to RS/6000. I
rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix). S/88 Product
Administrator started taking us around to their customers and also had
me write a section for the corporate continuous availability
document (it gets pulled when both AS400/Rochester and mainframe/POK
complain they couldn't meet requirements). Also previously worked on
original SQL/relational, System/R with Jim Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on anything with more than 4-system clusters, then
leave IBM a few months later.
Had left IBM and was brought in as consultant to small client/server
startup by two former Oracle employees (worked with on RDBMS and were
in the Ellison/Hester meeting) that were there responsible for
something called "commerce server" and they wanted to do payment
transactions on the server. The startup had also invented this
technology called SSL they wanted to use, it is now sometimes called
"electronic commerce". I had responsibility for everything between
webservers and payment networks. Based on procedures, documentation
and software I had to do for electronic commerce, I did a talk on "Why
Internet Wasn't Business Critical Dataprocessing" that (Internet/IETF
standards editor) Postel sponsored at ISI/USC.
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmpcontinuous availability, disaster survivability, geographic
survivability posts
https://www.garlic.com/~lynn/submain.html#available
Original SQL/relational, System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
Internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
electronic commerce gateway posts
https://www.garlic.com/~lynn/subnetwork.html#gateway
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Mainframe TCP/IP and RFC1044
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Mainframe TCP/IP and RFC1044
Date: 15 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#18 IBM Mainframe TCP/IP and RFC1044
https://www.garlic.com/~lynn/2025e.html#19 IBM Mainframe TCP/IP and RFC1044
yes, NSC RFC1044. I got sucked into doing the NSC channel extender
support first for IBM STL. In 1980 ... STL was bursting at the seams
and they were moving 300 people from STL IMS group to offsite bldg
with dataprocessing back to STL. They had tried "remote 3270" and
found human factors totally unacceptable. I then implemented the
channel extender support for NSC and they found no perceptible
difference between offsite and inside STL. An unanticipated
side-effect was it improved system throughput by 10-15%. STL had
spread the 3270 controllers across same channels with 3830 disk
controllers ... the NSC channel extender significantly reduced the
channel busy for the same amount of 3270 activity ... improving disk
and overall system throughput. There was some consideration of then
configuring all STL systems with channel extender (even for 3270s
inside STL).
Then NSC tried to get IBM to release my support, but there were some
people in POK working on serial and they got it vetoed (because they
were worried it might affect justify the release of their serial
stuff.
1988, the branch office asks if I could help LLNL (national lab)
standardize some serial they were working with, which quickly becomes
fibre-channel standard, "FCS" (not First Customer Ship), initially
1gbit/sec transfer, full-duplex, aggregate 200mbye/sec (including some
stuff I had done in 1980). Then POK finally gets their serial stuff
released with ES/9000 as ESCON (when it was already obsolete,
initially 10mbytes/sec, later increased to 17mbytes/sec). Then some
POK engineers become involved with FCS and define a heavy-weight
protocol that drastically cuts throughput (eventually released as
FICON). 2010, a z196 "Peak I/O" benchmark released, getting 2M IOPS
using 104 FICON (20K IOPS/FICON). About the same time a FCS is
announced for E5-2600 server blade claiming over million IOPS (two
such FCS having higher throughput than 104 FICON). Also IBM pubs
recommend that SAPs (system assist processors that actually do I/O) be
kept to 70% CPU (or 1.5M IOPS) and no new CKD DASD has been made for
decades, all being simulated on industry standard fixed-block devices.
channel extender posts
https://www.garlic.com/~lynn/submisc.html#channel.extender
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
--
virtualization experience starting Jan1968, online at home since Mar1970
Opel
From: Lynn Wheeler <lynn@garlic.com>
Subject: Opel
Date: 16 Oct, 2025
Blog: Facebook
related; 1972, Learson tried (and failed) to block bureaucrats,
careerists, and MBAs from destroying Watson culture/legacy, pg160-163,
30yrs of management briefings 1958-1988
https://bitsavers.org/pdf/ibm/generalInfo/IBM_Thirty_Years_of_Mangement_Briefings_1958-1988.pdf
Early/mid 70s, was IBM's Future System; FS was totally different from
370 and was going to completely replace it. During FS, internal
politics was killing off 370 projects and lack of new 370 is credited
with giving the clone 370 makers (including Amdahl), their market
foothold.
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
F/S implosion, from 1993 Computer Wars: The Post-IBM World
https://www.amazon.com/Computer-Wars-The-Post-IBM-World/dp/1587981394/
... and perhaps most damaging, the old culture under Watson Snr and Jr
of free and vigorous debate was replaced with *SYNCOPHANCY* and *MAKE
NO WAVES* under Opel and Akers. It's claimed that thereafter, IBM
lived in the shadow of defeat ... But because of the heavy investment
of face by the top management, F/S took years to kill, although its
wrong headedness was obvious from the very outset. "For the first
time, during F/S, outspoken criticism became politically dangerous,"
recalls a former top executive
... snip ...
20yrs after Learson's failure, IBM has one of the largest losses in
the history of US companies. IBM was being reorganized into the 13
"baby blues" in preparation for breaking up the company (take off on
"baby bells" breakup decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist
some more from token-ring post yesterday in (public) mainframe group:
https://www.garlic.com/~lynn/2025e.html#23 IBM Token-Ring
https://www.garlic.com/~lynn/2025e.html#22 IBM Token-Ring
https://www.garlic.com/~lynn/2025e.html#21 IBM Token-Ring
senior disk engineer gets talk scheduled at annual, internal,
world-wide communication group conference, supposedly on 3174
performance. However, his opening was that the communication group was
going to be responsible for the demise of the disk division. The disk
division was seeing drop in disk sales with data fleeing mainframe
datacenters to more distributed computing friendly platforms. The disk
division had come up with a number of solutions, but they were
constantly being vetoed by the communication group (with their
corporate ownership of everything that crossed the datacenter walls)
trying to protect their dumb terminal paradigm.
Senior disk software executive partial countermeasure was investing in
distributed computing startups that would use IBM disks (he would
periodically ask us to drop in on his investments to see if we could
offer any assistance). The communication group stranglehold on
mainframe datacenters wasn't just disk and a couple years later, IBM
has one of the largest losses in the history of US companies
future system posts
https://www.garlic.com/~lynn/submain.html#futuresys
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Germany
Date: 18 Oct, 2025
Blog: Facebook
When I graduated and joined the IBM Cambridge Science Center, one of
my hobbies was enhanced production operating systems for internal
datacenters and (online branch office sales&marketing support) US HONE
systems was one of the first (and long time) customer. Then got one of
my overseas business trips to both Paris, for 1st non-US HONE install
and to Boeblingen (put me up in small business traveler's hotel in
residential district and hotels changed four times the telco tariff
(like $60).
During FS in the early 70s, internal politics were killing 370 efforts
and the lack of new 370 is credited with giving the clone 370 makers
their market foothold). After Future System implodes, there is mad
rush to get stuff back into the 370 product pipelines (including
quick&dirty 3033&3081 efforts)
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
About the same time Endicott cons me into helping with 138/148 ECPS
microcode assist and was also con'ed into helping with a 125-II five
CPU implementation (115 has nine microprocessor memory bus, all the
microprocessors the same, including 370, just different microcode
loads; 125 is identical except the microprocessor running 370
microcode is 50% faster). Endicott then complains that the 5-CPU 125
would overlap the throughput of 148/ECPS ... and at escalation
meeting, I had to argue both sides of the table (but the 125 5-CPU
gets shutdown).
Later in 70s I had transferred to SJR on the west coast and get to
wander around IBM (and non-IBM) datacenters in silicon valley,
including disk bldg14/engineering and bldg15/product test across the
street. They were running 7x24, pre-scheduled, stand-alone testing and
mentioned that they had recently tried MVS, but it had 15min MTBF (in
that environment) requiring manual re-ipl. I offer to rewrite I/O
system to make it bullet proof and never fail, allowing any amount of
on-demand testing, greatly improving productivity. Bldg15 then got 1st
engineering 3033 (1st outside POK processor engineering) and since I/O
testing only used a percent or two of CPU, we scrounge up a 3830 and a
3330 string for our own, private online service. At the time, air
bearing simulation (for thin-film head design) was only getting a
couple turn arounds a month on SJR 370/195. We set it up on bldg15
3033 (slightly less than half 195 MIPS) and they could get several
turn arounds a day. Used in 3370FBA then in 3380CKD
https://www.computerhistory.org/storageengine/thin-film-heads-introduced-for-large-disks/
trivia: 3380 was already transitioning to fixed-block, can be seen in
the records/track formulas where record size had to be rounded up to
multiple of fixed "cell" size.
science center posts
https://www.garlic.com/~lynn/subtopic.html#545tech
cp/67l, csc/vm, sjr/vm posts
https://www.garlic.com/~lynn/submisc.html#cscvm
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
future system posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP, tightly-coupled, shared-memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
125 5-cpu project
https://www.garlic.com/~lynn/submain.html#bounce
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Thin Film Disk Head
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Thin Film Disk Head
Date: 19 Oct, 2025
Blog: Facebook
2nd half of 70s transferring to SJR on west coast, I worked with Jim
Gray and Vera Watson on original SQL/relational, System/R (done on
VM370); we were able to do tech transfer ("under the radar" while
corporation was pre-occupied with "EAGLE") to Endicott for
SQL/DS. Then when "EAGLE" implodes, there was request for how fast
could System/R be ported to MVS ... which was eventually released as
DB2, originally for decision-support *only*.
I also got to wander around IBM (and non-IBM) datacenters in silicon
valley, including DISK bldg14 (engineering) and bldg15 (product test)
across the street. They were running pre-scheduled, 7x24, stand-alone
testing and had mentioned recently trying MVS, but it had 15min MTBF
(requiring manual re-ipl) in that environment. I offer to redo I/O
subsystem to make it bullet proof and never fail, allowing any amount
of on-demand testing, greatly improving productivity. Bldg15 then gets
1st engineering 3033 outside POK processor engineering ... and since
testing only took percent or two of CPU, we scrounge up 3830
controller and 3330 string to setup our own private online service.
Air bearing simulation (for thin film heads) was getting a couple turn
arounds a month on the SJR MVT 370/195. We set it up on the bldg15
3033 and it was able to get as many turn arounds a day as they wanted.
Then bldg15 also gets engineering 4341 in 1978 and some how branch
hears about it and in Jan1979 I'm con'ed into doing a 4341 benchmark
for a national lab that was looking at getting 70 for compute farm
(leading edge of the coming cluster supercomputing tsunami).
first thin film head was 3370
https://www.computerhistory.org/storageengine/thin-film-heads-introduced-for-large-disks/
then used for 3380; original 3380 had 20 track spacings between each
data track, then cut the spacing in half for double the capacity, then
cut the spacing again for triple the capacity (3380K). The "father of
risc" then talks me into helping with a "wide" disk head design,
read/write 16 closely spaced data tracks in parallel (plus two servo
tracks, one on each side of 16 data track groupings). Problem was data
rate would have been 50mbytes/sec at a time when mainframe channels
were still 3mbytes/sec. However 40mbyte/sec disk arrays were becoming
common and Cray channel had been standardized as HIPPI (100mbyte/sec)
https://en.wikipedia.org/wiki/HIPPI
1988, IBM branch asks if I could help LLNL (national lab) standardize
some serial stuff they were working with, which quickly becomes fibre
channel standard ("FCS", initially 1gbit, full-duplex, got RS/6000
cards capable of 200mbytes/sec aggregate for use with 64-port FCS
switch). In 1990s, some serial stuff that POK had been working with
for at least the previous decade is released as ESCON (when it is
already obsolete, 10mbytes/sec, upgraded to 17mbytes/sec). Then some
POK engineers become involved with FCS and define heavy weight
protocol that significantly reduces ("native") throughput, which ships
as "FICON".
Latest public benchmark I've seen was z196 "Peak I/O" getting 2M IOPS
using 104 FICON (20K IOPS/FICON). About the same time a FCS is
announced for E5-2600 blades claiming over a million IOPS (two such
FCS having higher throughput than 104 FICON). Also IBM pubs
recommended that SAPs (system assist processors that do actual I/O) be
held to 70% CPU (or around 1.5M IOPS) and no CKD DASD have been made
for decades, all being simulated on industry standard fixed-block
devices. https://en.wikipedia.org/wiki/Fibre_Channel
posts mentioning working with Jim Gray and Vera Watson on original SQL/relational, System/R
https://www.garlic.com/~lynn/submain.html#systemr
posts mentioning getting to work in bldg14&15
https://www.garlic.com/~lynn/subtopic.html#disk
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Germany
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Germany
Date: 20 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#28 IBM Germany
After 125 5-cpu got shutdown was asked to help with a 16-cpu 370 and
we con the 3033 processor engineers into helping with it in their
spare time (a lot more interesting than remapping 168 logic to 20%
faster chips). Everybody thought it was great until somebody tells the
head of POK that it could be decades until POK's favorite son
operating system ("MVS") had (effective) 16-cpu support (POK doesn't
ship a 16-CPU system until after the turn of the century). At the
time, MVS documents had 2-CPU support only getting 1.2-1.5 the times
the throughput of 1-CPU (because of the heavy weight multiprocessor
support). Head of POK then invites some of us to never visit POK
again and instructs the 3033 processor engineers keep heads down and
no distractions.
Contributing was head of POK was in the process of convincing
corporate to kill the VM370 product, shutdown the development group
and transfer all the people to POK for MVS/XA. Endicott eventually
manages to save the VM370 product mission (for the mid-range), but has
to recreate a development group from scratch.
SMP, tightly-coupled, shared-memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: What Is A Mainframe
Date: 21 Oct, 2025
Blog: Facebook
Early numbers actual industry benchmarks (number of program iterations
compared to industry standard MIPS/BIPS reference platform), later
used IBM pubs giving percent change since previous
z900, 16 processors 2.5BIPS (156MIPS/core), Dec2000
z990, 32 cores, 9BIPS, (281MIPS/core), 2003
z9, 54 cores, 18BIPS (333MIPS/core), July2005
z10, 64 cores, 30BIPS (469MIPS/core), Feb2008
z196, 80 cores, 50BIPS (625MIPS/core), Jul2010
EC12, 101 cores, 75BIPS (743MIPS/core), Aug2012
z13, 140 cores, 100BIPS (710MIPS/core), Jan2015
z14, 170 cores, 150BIPS (862MIPS/core), Aug2017
z15, 190 cores, 190BIPS (1000MIPS/core), Sep2019
z16, 200 cores, 222BIPS (1111MIPS/core), Sep2022
z17, 208 cores, 260BIPS (1250MIPS/core), Jun2025
In 1988, the IBM branch office asks if I could help LLNL (national
lab) standardize some serial they were working with, which quickly
becomes fibre-channel standard, "FCS" (not First Customer Ship),
initially 1gbit/sec transfer, full-duplex, aggregate 200mbye/sec
(including some stuff I had done in 1980).
Then POK gets some of their serial stuff released with ES/9000 as
ESCON (when it was already obsolete, initially 10mbytes/sec, later
increased to 17mbytes/sec). Then some POK engineers become involved
with FCS and define a heavy-weight protocol that drastically cuts
throughput (eventually released as FICON). 2010, a z196 "Peak I/O"
benchmark released, getting 2M IOPS using 104 FICON (20K
IOPS/FICON). About the same time a FCS is announced for E5-2600 server
blade claiming over million IOPS (two such FCS having higher
throughput than 104 FICON). Also IBM pubs recommend that SAPs (system
assist processors that actually do I/O) be kept to 70% CPU (or 1.5M
IOPS) and no new CKD DASD has been made for decades, all being
simulated on industry standard fixed-block devices.
Note: 2010 E5-2600 server blade (16 cores, 31BIPS/core) benchmarked at
500BIPS (ten times max configured Z196). At the time, commonly used in
cloud megadatacenters (each having half million or more server blades)
trivia: in the wake of the Future System implosion in the mid-70s,
there is mad rush to get stuff back into the 370 product pipelines and
I get asked to help with a 16-CPU 370 effort and we con the 3033
processor (started out remapping 168 logic to 20% faster chips)
engineers into working on it in their spare time (lot more interesting
than the 168 logic remapping). Everybody thought it was great until
somebody tells head of POK (IBM high-end 370) that it could be decades
before POK's favorite son operating system ("MVS") had ("effective")
16-CPU support (MVS docs had 2-CPU throughput only 1.2-1.5 times
throughput of single CPU because of its high-overhead multiprocessor
support, POK doesn't ship 16-CPU system until after turn of
century). Head of POK then invites some of us to never visit POK again
and directs the 3033 processor engineers heads down and no
distractions.
Also 1988, Nick Donofrio approves HA/6000, originally for NYTimes to
move their newspaper system (ATEX) off VAXCluster to RS/6000. I rename
it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Ingres, Sybase, Informix that have DEC
VAXCluster support in same source base with UNIX). IBM S/88 Product
Administrator was also taking us around to their customers and also
had me write a section for corporate continuous availability strategy
document (it gets pulled when both Rochester/AS400 and POK/mainframe
complain).
Early Jan92 meeting with Oracle CEO, AWD executive Hester tells
Ellison that we would have 16-system clusters by mid92 and 128-system
clusters by ye92. Mid-jan92 convince FSD to bid HA/CMP for
gov. supercomputers. Late-jan92, cluster scale-up is transferred for
announce as IBM Supercomputer (for technical/scientific *only*) and we
were told we couldn't work on clusters with more than four systems (we
leave IBM a few months later).
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to the
industry MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Former executive we had reported to, goes over to head up Somerset/AIM
(Apple, IBM, Motorola), single chip RISC with M88k bus/cache (enabling
clusters of shared memory multiprocessors)
i86 chip makers then do hardware layer that translate i86 instructions
into RISC micro-ops for actual execution (largely negating throughput
difference between RISC and i86); 1999 industry benchmark:
IBM PowerPC 440: 1,000MIPS
Pentium3: 2,054MIPS (twice PowerPC 440)
Fibre Channel Standard ("FCS") &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
SMP, tightly-coupled, shared-memory multiprocessor
https://www.garlic.com/~lynn/subtopic.html#smp
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
posts mentioning availability
https://www.garlic.com/~lynn/submain.html#available
posts mentioning assurance
https://www.garlic.com/~lynn/subintegrity.html#assurance
DASD, CKD, FBA, multi-track posts
https://www.garlic.com/~lynn/submain.html#dasd
--
virtualization experience starting Jan1968, online at home since Mar1970
What Is A Mainframe
From: Lynn Wheeler <lynn@garlic.com>
Subject: What Is A Mainframe
Date: 21 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#32 What Is A Mainframe
During FS period, internal politics was killing off 370 efforts
(claims that lack of new 370s during FS gave the 370 clone makers
their market foothold).
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
when FS implodes there was mad rush to get stuff back into the 370
product pipelines, including kicking off quick&dirty 3033&3081
efforts in parallel. Original 3081D (2-CPU) aggregate MIPS was less
than Amdahl 1-CPU system. Then IBM doubles the processors' cache size,
making 3081K 2-CPU about the same MIPS as Amdahl 1-CPU (modulo 3081K
2-CPU MVS throughput was only .6-.75 the throughput of Amdahl 1-CPU,
because of its significant multiprocessor overhead). Then because
ACP/TPF didn't have SMP, tightly-coupled, multiprocessor support,
there was concern that the whole ACP/TPF market would move to
Amdahl. The 3081 2nd CPU was in the middle of the box, concern just
removing it (for 1-CPU 3083) would make it top heavy and prone to
falling over (so box had to be rewired to move CPU0 to the middle)
trivia: before MS/DOS:
https://en.wikipedia.org/wiki/MS-DOS
there was Seattle computer
https://en.wikipedia.org/wiki/Seattle_Computer_Products
before Seattle computer, there was CP/M (name take-off on IBM CP/67)
https://en.wikipedia.org/wiki/CP/M
before developing CP/M, kildall worked on IBM CP/67 at npg
https://en.wikipedia.org/wiki/Naval_Postgraduate_School
... more trivia:
I took two credit hr, intro to fortran/computers and at end of
semester, I was hired to rewrite 1401 MPIO for 360/30. Univ. was
getting 360/67 for tss/360 replacing 709/1401 and 360/30 temporarily
replacing 1401 pending availability of 360/67 ... and within few weeks
had 2000 card 360 assembler program (univ. shutdown datacenter on
weekends and I had the datacenter dedicated, although 48hrs w/o sleep
made monday classes hard). Within a year of taking intro class and I
was hired fulltime responsible for OS/360 (360/67 running 360/65,
tss/360 didn't come to production). Jan, 1968, CSC came out to install
CP/67 (precursor to VM370), 3rd installation after CSC itself and MIT
Lincoln Lab. I mostly got to play with it during my dedicated weekend
time, initially working on pathlengths for running OS/360
in virtual machine. OS/360 benchmark ran 322secs on real
machine, initially 856secs in virtual machine (CP67 CPU
534secs), after a couple months I have reduced that CP67 CPU from
534secs to 113secs. I then start rewriting the dispatcher, (dynamic
adaptive resource manager/default fair share policy) scheduler,
paging, adding ordered seek queuing (from FIFO) and mutli-page
transfer channel programs (from FIFO and optimized for
transfers/revolution, getting 2301 paging drum from 70-80 4k
transfers/sec to channel transfer peak of 270). Six months after univ
initial install, CSC was giving one week class in LA. I arrive on
Sunday afternoon and asked to teach the class, it turns out that the
people that were going to teach it had resigned the Friday before to
join one of the 60s CP67 commercial online spin-offs.
Originally CP67 delivered with 1052&2741 terminal support and
automagic terminal type (switching terminal type port scanner with SAD
CCW). Univ. had some ASCII TTYs and I add TTY integrated with
automagic termainl type. I then wanted to have single terminal dial-in
phone number ("hunt group"), but it didn't work since IBM had taken
short cut and hard-wired line-speed for each port. This kicks off univ
clone controller effort, build channel interface board for Interdata/3
programmed to emulate IBM controller with addition of auto-baud
terminal line support. This is upgraded to Interdata/4 for channel
interface and clusters of Interdata/3s for ports. Interdata (and then
Perrkin-Elmer) sells it as clone controller and four of use are
written up as responsible for (some part of) the clone controller
business
https://en.wikipedia.org/wiki/Interdatahttps://en.wikipedia.org/wiki/Perkin-Elmer#Computer_Systems_Division
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
CP67l, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
clone controller posts
https://www.garlic.com/~lynn/submain.html#360pcm
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Linux Clusters
Date: 21 Oct, 2025
Blog: Facebook
basically cluster supercomputers and cloud clusters are similar with
large numbers of linux servers all tied together. recent article the
top 500 supercomputers are all linux clusters. a large cloud operation
can have a score or more of megadatacenters around the world, each
megadatacenter with half million or more of linux server blades and
each server blade with ten times the processing of max configured
mainframe (and enormous automation, a megadatacenter run with 70-80
staff). Decade ago there were articles of being able to use credit
card at a cloud operation to automagically spin up blades for a 3hr
supercomputer ranking in the top 40 in the world.
trivia: 1988, IBM branch asks if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre channel standard ("FCS", initially 1gbit, full-duplex,
got RS/6000 cards capable of 200mbytes/sec aggregate for use with
64-port FCS switch). In 1990s, some serial stuff that POK had been
working with for at least the previous decade is released as ESCON
(when it is already obsolete, 10mbytes/sec, upgraded to
17mbytes/sec). Then some POK engineers become involved with FCS and
define heavy weight protocol that significantly reduces ("native")
throughput, which ships as "FICON". Latest public benchmark I've seen
was z196 "Peak I/O" getting 2M IOPS using 104 FICON (20K
IOPS/FICON). About the same time a FCS is announced for E5-2600 server
blades claiming over a million IOPS (two such FCS having higher
throughput than 104 FICON). Also IBM pubs recommended that SAPs
(system assist processors that do actual I/O) be held to 70% CPU (or
around 1.5M IOPS) and no CKD DASD have been made for decades, all
being simulated on industry standard fixed-block devices..
The max configured z196 benchmarked at 50BIPS and went for $30M. An
E5-2600 server blade benchmarked at 500BIPS (ten times z196, same
industry benchmark, number of program iterations compared to industry
MIPS/BIPS reference platform) and had a IBM base list price of $1815
(shortly later industry pubs had blade server component makers
shipping half their products directly to cloud operations that
assemble their own servers at 1/3rd the cost of brand name servers
... and IBM sells off its server business).
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
megadatacenter poss
https://www.garlic.com/~lynn/submisc.html#megadatacenter
DASD, CKD, FBA, multi-track search posts
https://www.garlic.com/~lynn/submain.html#dasd
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Linux Clusters
Date: 24 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#34 Linux Clusters
https://www.garlic.com/~lynn/2025e.html#35 Linux Clusters
In the 70s, with the implosion of Future System (internal politics had
been killing off 370 efforts, claim is the lack of new 370s during FS
is credited with giving clone 370 makers their market foothold), there
was mad rush to get stuff back into the 370 product pipelines,
including kicking off quick&dirty 3033&3081 efforts in parrallel.
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
I get asked to help with a 16-CPU 370, and we con the 3033 processor
engineers into helping in their spare time (a lot more interesting
than remapping 168 logic to 20% faster chips). Everybody thought it
was great until somebody tells the head of POK that it could be
decades before POK favorite son operating system ("MVS") had
(effective) 16-CPU support (MVS docs at the time saying 2-CPU systems
only had 1.2-1.5 times throughput of single CPU because of heavy SMP
overhead, aka MVS 2-CPU 3081K at same aggregate MIPS as Amdahl single
processor, only had .6-.75 the throughput); POK doesn't ship 16-CPU
system until after turn of century. Then head of POK invites some of
us to never visit POK again and directs 3033 processor engineers heads
down and no distractions.
future system posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP. tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
--
virtualization experience starting Jan1968, online at home since Mar1970
Linux Clusters
From: Lynn Wheeler <lynn@garlic.com>
Subject: Linux Clusters
Date: 24 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#34 Linux Clusters
https://www.garlic.com/~lynn/2025e.html#35 Linux Clusters
https://www.garlic.com/~lynn/2025e.html#36 Linux Clusters
related; 1972, Learson tried (and failed) to block bureaucrats,
careerists, and MBAs from destroying Watson culture/legacy, pg160-163,
30yrs of management briefings 1958-1988
https://bitsavers.org/pdf/ibm/generalInfo/IBM_Thirty_Years_of_Mangement_Briefings_1958-1988.pdf
Early/mid 70s, was IBM's Future System; FS was totally different from
370 and was going to completely replace it. During FS, internal
politics was killing off 370 projects and lack of new 370 is credited
with giving the clone 370 makers (included Amdahl), their market
foothold.
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
F/S implosion, from 1993 Computer Wars: The Post-IBM World
https://www.amazon.com/Computer-Wars-The-Post-IBM-World/dp/1587981394/
... and perhaps most damaging, the old culture under Watson Snr and Jr
of free and vigorous debate was replaced with *SYNCOPHANCY* and *MAKE
NO WAVES* under Opel and Akers. It's claimed that thereafter, IBM
lived in the shadow of defeat ... But because of the heavy investment
of face by the top management, F/S took years to kill, although its
wrong headedness was obvious from the very outset. "For the first
time, during F/S, outspoken criticism became politically dangerous,"
recalls a former top executive
... snip ...
Late 80s, a senior disk engineer gets talk scheduled at annual,
internal, world-wide communication group conference, supposedly on
3174 performance. However, his opening was that the communication
group was going to be responsible for the demise of the disk
division. The disk division was seeing drop in disk sales with data
fleeing mainframe datacenters to more distributed computing friendly
platforms. The disk division had come up with a number of solutions,
but they were constantly being vetoed by the communication group (with
their corporate ownership of everything that crossed the datacenter
walls) trying to protect their dumb terminal paradigm. Senior disk
software executive partial countermeasure was investing in
distributed computing startups that would use IBM disks (he would
periodically ask us to drop in on his investments to see if we could
offer any assistance).
The communication group stranglehold on mainframe datacenters wasn't
just disk and a couple years later (20yrs after Learson's failure),
IBM has one of the largest losses in the history of US companies, and
was being reorganized into the 13 "baby blues" in preparation for
breaking up the company (take off on "baby bells" breakup decade
earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
future system posts
https://www.garlic.com/~lynn/submain.html#futuresys
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
--
virtualization experience starting Jan1968, online at home since Mar1970
Amazon Explains How Its AWS Outage Took Down the Web
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Boca and IBM/PCs
Date: 25 Oct, 2025
Blog: Facebook
30yrs of PC market ("IBM" PCs increasingly dominated by "clones")
https://arstechnica.com/features/2005/12/total-share/
I had been posting to the internal PC forums, quantity one prices for
clones advertised in Sunday SJMN ... totally different from Boca
projections. Head of POK mainframe is moved to Boca to head up
PCs. They contract with Dataquest (since bought by Gartner) for study
of the PC market future, including a video taped roundtable of Silicon
Valley experts. I'd known the Dataquest person for a decade, and was
asked to be a Silicon Valley "expert" (promising that they would
garble my identity so Boca wouldn't recognize me as an IBM employee),
and after clearing it with my IBM management, and agreed to
participate.
trivia: Late 70s and early 80s, I had been blamed for online computer
conferencing on the internal network (precursor to social media,
larger than arpanet/internet from just about the beginning until
sometime mid/late 80s --- about the time it was forced to convert to
SNA/VTAM). Only about 300 actually participated but claims 25,000 were
reading. When the corporate executive committee was told about it,
folklore is 5of6 wanted to fire me. Results included officially
sanctioned software & forums and researcher paid to sit in back of
my office for nine months studying how I communicated, face-to-face
& telephone; all incoming & outgoing email, instant messages
logs (results were IBM reports, papers and conference talks, books,
and a Stanford Phd joint with language and computer AI).
AWD (workstation division) had done their own cards for PC/RT (16bit
PC/AT bus) ... including 4mbit Token-Ring card. For the RS/6000
(w/microchannel, 32bit bus), AWD was told they couldn't do their own
cards, but had to use the (communication group heavily performance
kneecapped) PS2 cards. It turns out the PS2 microchannel 16mbit
Token-Ring card had lower throughput than the PC/RT 4mbit Token-Ring
card (i.e. joke that PC/RT server with 4mbit T/R would have higher
throughput than RS/6000 server with 16mbit T/R).
New Almaden Research bldg had been heavily provisioned with CAT wiring
assuming 16mbit token-ring, but found 10mbit ethernet (over CAT
wiring) LAN had lower latency and higher aggregate throughput than
16mbit token-ring LAN. They also found $69 10mbit ethernet cards had
higher throughput than the PS2 microchannel $800 16mbit token-ring
cards. We were out giving customer executive presentations on TCP/IP,
10mbit ethernet, 3-tier architecture, high-performance routers,
distributed computing presentations (including comparisons with
standard IBM offerings) and taking misinformation barbs in the back by
the SNA, SAA, & token-ring forces. The Dallas E&S center
published something purported to be 16mbit T/R compared to Ethernet
... but only (remotely valid) explanation I could give was that they
compared to early 3mbit ethernet prototype predating
listen-before-transmit (CSMA/CD) part of Ethernet protocol standard.
About the same time, senior disk engineer gets talk scheduled at
annual, internal, world-wide communication group conference,
supposedly on 3174 performance. However, his opening was that the
communication group was going to be responsible for the demise of the
disk division. The disk division was seeing drop in disk sales with
data fleeing mainframe datacenters to more distributed computing
friendly platforms. The disk division had come up with a number of
solutions, but they were constantly being vetoed by the communication
group (with their corporate ownership of everything that crossed the
datacenter walls) trying to protect their dumb terminal paradigm.
Senior disk software executive partial countermeasure was
investing in distributed computing startups that would use IBM disks
(he would periodically ask us to drop in on his investments to see if
we could offer any assistance). The communication group stranglehold
on mainframe datacenters wasn't just disk and a couple years later,
IBM has one of the largest losses in the history of US companies and
was being reorganized into the 13 "baby blues" in preparation for
breaking up the company ("baby blues" take-off on the "baby bell"
breakup decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
online computer conferencing posts
https://www.garlic.com/~lynn/subnetwork.html#cmc
IBM communication group predicted to be responsible for demise of disk
division
https://www.garlic.com/~lynn/subnetwork.html#emulation
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 360/85
Date: 26 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#41 IBM 360/85
Amdahl wins battle to make ACS, 360-compatible. Then ACS/360 is killed
(folklore was concern that it would advance state-of-the-art too fast)
and Amdahl then leaves IBM (before Future System); end ACS/360:
https://people.computing.clemson.edu/~mark/acs_end.html
Besides hobby of doing enhanced production operating systems for
internal datacenters ... and wandering around internal datacenters
... I spent some amount of time at user group meetings (like SHARE)
and wandering around customers. Director of one of the largest
(customer) financial datacenters liked me to drop in and talk
technology. At one point, the branch manager horribly offended the
customer and in retaliation, they ordered an Amdahl machine (lonely
Amdahl clone 370 in a vast sea of "blue). Up until then Amdahl had
been selling into univ. & tech/scientific markets, but clone 370s had
yet to break into the IBM true-blue commercial market ... and this
would be the first. I got asked to go spend 6m-12m on site at the
customer (to help obfuscate the reason for the Amdahl order?). I
talked it over with the customer, who said while he would like to have
me there it would have no affect on the decision, so I declined the
offer. I was then told the branch manager was good sailing buddy of
IBM CEO and I could forget a career, promotions, raises.
Early 70s, there was Future System project (and internal politics was
killing off 370 efforts, claim is that lack of new 370s during FS
contributed to clone 370 makers their market foothold)
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
When FS implodes there was mad rush to get stuff back into the 370
product pipelines, including kicking off quick&dirty 3033&3081 efforts
in parallel and I got asked to help with a 16-CPU 370, and we con the
3033 processor engineers into helping in their spare time (a lot more
interesting than remapping 168 logic to 20% faster chips). Everybody
thought it was great until somebody tells the head of POK that it
could be decades before POK favorite son operating system ("MVS") had
(effective) 16-CPU support (MVS docs at the time saying 2-CPU systems
only had 1.2-1.5 times throughput of single CPU because of heavy SMP
overhead). Then head of POK invites some of us to never visit POK
again and directs 3033 processor engineers heads down and no
distractions.
Original 3081D (2-CPU) aggregate MIPS was less than Amdahl 1-CPU
system. Then IBM doubles the processors' cache size, making 3081K
2-CPU about the same MIPS as Amdahl 1-CPU. MVS 2-CPU 3081K at same
aggregate MIPS as Amdahl single processor, only had .6-.75 the
throughput (because of MVS multiprocessor overhead). POK doesn't ship
16-CPU system until after turn of century.
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
some ACS/360 and Amdahl clone 370s posts
https://www.garlic.com/~lynn/2025e.html#16 CTSS, Multics, Unix, CSC
https://www.garlic.com/~lynn/2025d.html#99 IBM Fortran
https://www.garlic.com/~lynn/2025d.html#61 Amdahl Leaves IBM
https://www.garlic.com/~lynn/2025c.html#49 IBM And Amdahl Mainframe
https://www.garlic.com/~lynn/2024f.html#122 IBM Downturn and Downfall
https://www.garlic.com/~lynn/2024f.html#23 Future System, Single-Level-Store, S/38
https://www.garlic.com/~lynn/2022g.html#59 Stanford SLAC (and BAYBUNCH)
https://www.garlic.com/~lynn/2022b.html#88 Computer BUNCH
https://www.garlic.com/~lynn/2022.html#74 165/168/3033 & 370 virtual memory
https://www.garlic.com/~lynn/2021k.html#105 IBM Future System
https://www.garlic.com/~lynn/2021e.html#66 Amdahl
https://www.garlic.com/~lynn/2021.html#39 IBM Tech
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 360/85
Date: 26 Oct, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#41 IBM 360/85
https://www.garlic.com/~lynn/2025e.html#42 IBM 360/85
1970, HAWK, 30MIPS processor
Date: 05/12/81 13:46:19
To: wheeler
RE: Competiton for resources in IBM
Before Bob Tomasulo left to go to work for STC, he told me many
stories about IBM. Around 1970, there was a project called HAWK. It
was to be a 30 MIPS uniprocessor. Bob was finishing up on the 360/91,
and wanted to go work on HAWK as his next project. He was told by
management that "there were already too many good people working over
there, and they really couldn't afford to let another good person work
on it"! They forced him to work on another project that was more
"deserving" of his talents. Bob never forgave them for that.
I guess IBM feels that resources are to be spread thinly, and that no
single project can have lots of talent on it. Any project that has
lots of talent will be raided sooner or later.
... snip ... top of post, old email index
Amdahl had won battle to make ACS, 360-compatible. Then ACS/360 is
killed (folklore was concern that it would advance state-of-the-art
too fast) and Amdahl then leaves IBM (before Future System); end
ACS/360:
https://people.computing.clemson.edu/~mark/acs_end.html
Shortly after joining IBM, I was asked to help with adding
multithreading (see patent refs in ACS web page) to 370/195. 195 had
out-of-order execution, but no branch prediction (or speculative
execution) and conditional branches drained pipeline so most code only
ran at half rated speed. Adding another i-stream, simulating 2nd-CPU,
each running at half speed, could keep all the execution units running
at full-speed (modulo MVT&MVS 2-CPU multiprocessor only got 1.2-1.5
the throughput of single CPU; because of the multiprocessor
implementation). Then decision was made to add virtual memory to all
370s and it was decided that it wasn't partical to add virtual memory
to 370/195 (and the multithreading was killed).
Early 70s, there was Future System project (and internal politics was
killing off new 370 efforts, claim is that lack of new 370s during FS
contributed to giving clone 370 makers their market foothold)
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
HAWK may have been killed like Amdahl's ACS/360, or because of virtual
memory decision (like 370/195 multithreading), or because of Future
System (don't know).
Besides hobby of doing enhanced production operating systems for
internal datacenters ... and wandering around internal datacenters
... I spent some amount of time at user group meetings (like SHARE)
and wandering around customers. Director of one of the largest
(customer) financial datacenters liked me to drop in and talk
technology. At one point, the branch manager horribly offended the
customer and in retaliation, they ordered an Amdahl machine (lonely
Amdahl clone 370 in a vast sea of "blue). Up until then Amdahl had
been selling into univ. & tech/scientific markets, but clone 370s
had yet to break into the IBM true-blue commercial market ... and this
would be the first. I got asked to go spend 6m-12m on site at the
customer (to help obfuscate the reason for the Amdahl order?). I
talked it over with the customer, who said while he would like to have
me there it would have no affect on the decision, so I declined the
offer. I was then told the branch manager was good sailing buddy of
IBM CEO and I could forget a career, promotions, raises.
When FS implodes there was mad rush to get stuff back into the 370
product pipelines, including kicking off quick&dirty 3033&3081
efforts in parallel. I got asked to help with a 16-CPU 370, and we con
the 3033 processor engineers into helping in their spare time (a lot
more interesting than remapping 168 logic to 20% faster
chips). Everybody thought it was great until somebody tells the head
of POK that it could be decades before POK favorite son operating
system ("MVS") had (effective) 16-CPU support (MVS docs at the time
saying 2-CPU systems only had 1.2-1.5 times throughput of single CPU
because of heavy SMP overhead). Then head of POK invites some of us to
never visit POK again and directs 3033 processor engineers heads down
and no distractions.
Original 3081D (2-CPU) aggregate MIPS was less than Amdahl 1-CPU
system. Then IBM doubles the processors' cache size, making 3081K
2-CPU about the same MIPS as Amdahl 1-CPU. MVS 2-CPU 3081K at same
aggregate MIPS as Amdahl single processor, only had .6-.75 the
throughput (because of MVS multiprocessor overhead). POK doesn't ship
16-CPU system until after turn of century.
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
some posts mentioning 370/195 multithreading and virtual memory
https://www.garlic.com/~lynn/2025c.html#112 IBM Virtual Memory (360/67 and 370)
https://www.garlic.com/~lynn/2025c.html#79 IBM System/360
https://www.garlic.com/~lynn/2025b.html#35 3081, 370/XA, MVS/XA
https://www.garlic.com/~lynn/2025.html#32 IBM 3090
https://www.garlic.com/~lynn/2024g.html#110 IBM 370 Virtual Storage
https://www.garlic.com/~lynn/2024f.html#24 Future System, Single-Level-Store, S/38
https://www.garlic.com/~lynn/2024e.html#115 what's a mainframe, was is Vax addressing sane today
https://www.garlic.com/~lynn/2024d.html#101 Chipsandcheese article on the CDC6600
https://www.garlic.com/~lynn/2024d.html#66 360/65, 360/67, 360/75 750ns memory
https://www.garlic.com/~lynn/2024c.html#20 IBM Millicode
https://www.garlic.com/~lynn/2024.html#24 Tomasulo at IBM
https://www.garlic.com/~lynn/2023f.html#89 Vintage IBM 709
https://www.garlic.com/~lynn/2023e.html#100 CP/67, VM/370, VM/SP, VM/XA
https://www.garlic.com/~lynn/2023d.html#32 IBM 370/195
https://www.garlic.com/~lynn/2023d.html#20 IBM 360/195
https://www.garlic.com/~lynn/2023b.html#6 z/VM 50th - part 7
https://www.garlic.com/~lynn/2023b.html#0 IBM 370
https://www.garlic.com/~lynn/2022h.html#17 Arguments for a Sane Instruction Set Architecture--5 years later
https://www.garlic.com/~lynn/2022d.html#34 Retrotechtacular: The IBM System/360 Remembered
https://www.garlic.com/~lynn/2022d.html#22 COMPUTER HISTORY: REMEMBERING THE IBM SYSTEM/360 MAINFRAME, its Origin and Technology
https://www.garlic.com/~lynn/2022d.html#12 Computer Server Market
https://www.garlic.com/~lynn/2022b.html#51 IBM History
https://www.garlic.com/~lynn/2022.html#60 370/195
https://www.garlic.com/~lynn/2022.html#31 370/195
https://www.garlic.com/~lynn/2021d.html#28 IBM 370/195
https://www.garlic.com/~lynn/2019d.html#62 IBM 370/195
https://www.garlic.com/~lynn/2018b.html#80 BYTE Magazine Pentomino Article
https://www.garlic.com/~lynn/2017g.html#39 360/95
https://www.garlic.com/~lynn/2017c.html#26 Multitasking, together with OS operations
https://www.garlic.com/~lynn/2017.html#90 The ICL 2900
https://www.garlic.com/~lynn/2017.html#3 Is multiprocessing better then multithreading?
https://www.garlic.com/~lynn/2016h.html#7 "I used a real computer at home...and so will you" (Popular Science May 1967)
https://www.garlic.com/~lynn/2015h.html#110 Is there a source for detailed, instruction-level performance info?
https://www.garlic.com/~lynn/2015c.html#26 OT: Digital? Cloud? Modern And Cost-Effective? Surprise! It's The Mainframe - Forbes
https://www.garlic.com/~lynn/2012d.html#73 Execution Velocity
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM SQL/Relational
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM SQL/Relational
Date: 27 Oct, 2025
Blog: Facebook
2nd half of 70s transferring to SJR on west coast, I worked with Jim
Gray and Vera Watson on original SQL/relational, System/R (done on
VM370, Backus office just down the hall and Codds office was on flr
above). Jim refs:
https://jimgray.azurewebsites.net/https://en.wikipedia.org/wiki/Jim_Gray_(computer_scientist)
and Vera Watson
https://en.wikipedia.org/wiki/Vera_Watson
... System/R
https://en.wikipedia.org/wiki/IBM_System_R
We were able to tech transfer ("under the radar" while corporation was
pre-occupied with "EAGLE") to Endicott for SQL/DS. Then when "EAGLE"
implodes, there was request for how fast could System/R be ported to
MVS ... which was eventually released as DB2, originally for
decision-support *only*. I also got to wander around IBM (and non-IBM)
datacenters in silicon valley, including DISK bldg14 (engineering) and
bldg15 (product test) across the street. They were running
pre-scheduled, 7x24, stand-alone testing and had mentioned recently
trying MVS, but it had 15min MTBF (requiring manual re-ipl) in that
environment. I offer to redo I/O system to make it bullet proof and
never fail, allowing any amount of on-demand testing, greatly
improving productivity. Bldg15 then gets 1st engineering 3033 outside
POK processor engineering ... and since testing only took percent or
two of CPU, we scrounge up 3830 controller and 3330 string to setup
our own private online service. Then bldg15 also gets engineering 4341
in 1978 and some how branch hears about it and in Jan1979 I'm con'ed
into doing a 4341 benchmark for a national lab that was looking at
getting 70 for compute farm (leading edge of the coming cluster
supercomputing tsunami).
in 1988, Nick Donofrio approved HA/6000, originally for NYTimes to
transfer their newspaper system (ATEX) from DEC VAXCluster to
RS/6000. I rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scaleup with RDBMS
vendors (oracle, sybase, ingres, informix) that had VAXCluster in same
source base with Unix. I do a distributed lock manager supporting
VAXCluster semantics (and especially Oracle and Ingres have a lot of
input on significantly improving scale-up performance). S/88 Product
Administrator started taking us around to their customers and also had
me write a section for the corporate continuous availability document
(it gets pulled when both AS400/Rochester and mainframe/POK complain
they couldn't meet requirements).
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems. Then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to MIPS
reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Executive we reported to, had gone over to head-up Somerset/AIM
(Apple, IBM, Motorola) and does single chip Power/PC ... using
Motorola 88K bus&cache ... enabling multiprocessor configurations
... further beefing up clusters with multiple processors/system.
other info in this recent thread
https://www.garlic.com/~lynn/2025e.html#41 IBM 360/85
https://www.garlic.com/~lynn/2025e.html#42 IBM 360/85
https://www.garlic.com/~lynn/2025e.html#43 IBM 360/85
In-between System/R (plus getting to play disk engineer in
bldgs14&15) and HA/CMP ... early 80s, I got HSDT project, T1 and
faster computer links (both terrestrial and satellite) and battles
with CSG (60s, IBM had 2701 supporting T1, 70s with SNA/VTAM and
issues, links were caped at 56kbit ... and I had to mostly resort to
non-IBM hardware). Internal IBM network required link
encryptors and I hated what I had to pay for T1 encryptors and
faster ones were hard to find. I became involved with effort,
objective was to handle at least 6mbytes/sec and cost less than $100
to build. The corporate crypto group first claimed it seriously
weakened crypto standard and couldn't be used. It took me 3months to
figure out how to explain what was happening, rather than weaker, it
was much stronger ... it was hallow victory. I was then told that only
one organization in the world was allowed to use such crypto, I could
make as many as I wanted, but they all had to be sent to them. It was
when I realized there was three kinds of crypto: 1) the kind they
don't care about, 2) the kind you can't do, 3) the kind you can only
do for them.
Also was working with NSF director and was suppose to get $20M to
interconnect the NSF Supercomputer centers. Then congress cuts the
budget, some other things happened and eventually there was RFP
released (in part based on what we already had running). NSF 28Mar1986
Preliminary Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet
System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
some posts mentioning three kinds of crypto
https://www.garlic.com/~lynn/2024e.html#125 ARPANET, Internet, Internal Network and DES
https://www.garlic.com/~lynn/2024e.html#28 VMNETMAP
https://www.garlic.com/~lynn/2024d.html#75 Joe Biden Kicked Off the Encryption Wars
https://www.garlic.com/~lynn/2024b.html#36 Internet
https://www.garlic.com/~lynn/2023f.html#79 Vintage Mainframe XT/370
https://www.garlic.com/~lynn/2023b.html#57 Ethernet (& CAT5)
https://www.garlic.com/~lynn/2022d.html#73 WAIS. Z39.50
https://www.garlic.com/~lynn/2022d.html#29 Network Congestion
https://www.garlic.com/~lynn/2022b.html#109 Attackers exploit fundamental flaw in the web's security to steal $2 million in cryptocurrency
https://www.garlic.com/~lynn/2022.html#57 Computer Security
https://www.garlic.com/~lynn/2021e.html#75 WEB Security
https://www.garlic.com/~lynn/2021e.html#58 Hacking, Exploits and Vulnerabilities
https://www.garlic.com/~lynn/2021d.html#17 The Rise of the Internet
https://www.garlic.com/~lynn/2021c.html#70 IBM/BMI/MIB
https://www.garlic.com/~lynn/2021b.html#57 In the 1970s, Email Was Special
https://www.garlic.com/~lynn/2021b.html#22 IBM Recruiting
https://www.garlic.com/~lynn/2019b.html#23 Online Computer Conferencing
https://www.garlic.com/~lynn/2017c.html#69 ComputerWorld Says: Cobol plays major role in U.S. government breaches
https://www.garlic.com/~lynn/2016c.html#57 Institutional Memory and Two-factor Authentication
https://www.garlic.com/~lynn/2015c.html#85 On a lighter note, even the Holograms are demonstrating
https://www.garlic.com/~lynn/2014i.html#54 IBM Programmer Aptitude Test
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Think
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Think
Date: 28 Oct, 2025
Blog: Facebook
"think" related; 1972, Learson tried (and failed) to block
bureaucrats, careerists, and MBAs from destroying Watson
culture/legacy, pg160-163, 30yrs of management briefings 1958-1988
https://bitsavers.org/pdf/ibm/generalInfo/IBM_Thirty_Years_of_Mangement_Briefings_1958-1988.pdf
Early/mid 70s, was IBM's Future System; FS was totally different from
370 and was going to completely replace it. During FS, internal
politics was killing off 370 projects and lack of new 370 is credited
with giving the clone 370 makers (included Amdahl), their market
foothold.
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
F/S implosion, from 1993 Computer Wars: The Post-IBM World
https://www.amazon.com/Computer-Wars-The-Post-IBM-World/dp/1587981394/
... and perhaps most damaging, the old culture under Watson Snr and Jr
of free and vigorous debate was replaced with *SYNCOPHANCY* and
*MAKE NO WAVES* under Opel and Akers. It's claimed that
thereafter, IBM lived in the shadow of defeat ... But because of the
heavy investment of face by the top management, F/S took years to
kill, although its wrong headedness was obvious from the very
outset. "For the first time, during F/S, outspoken criticism became
politically dangerous," recalls a former top executive
... snip ...
20yrs after Learson's failure, IBM has one of the largest losses in
the history of US companies. IBM was being reorganized into the 13
"baby blues" in preparation for breaking up the company (take off on
"baby bells" breakup decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
other contribution, late 80s, senior disk engineer gets talk scheduled
at annual, internal, world-wide communication group conference,
supposedly on 3174 performance. However, his opening was that the
communication group was going to be responsible for the demise of the
disk division. The disk division was seeing drop in disk sales with
data fleeing mainframe datacenters to more distributed computing
friendly platforms. The disk division had come up with a number of
solutions, but they were constantly being vetoed by the communication
group (with their corporate ownership of everything that crossed the
datacenter walls) trying to protect their dumb terminal paradigm.
Senior disk software executive partial countermeasure was
investing in distributed computing startups that would use IBM disks
(he would periodically ask us to drop in on his investments to see if
we could offer any assistance). The communication group stranglehold
on mainframe datacenters wasn't just disk and a couple years later,
IBM has one of the largest losses in the history of US companies and
was being reorganized into the 13 "baby blues" in preparation for
breaking up the company ("baby blues" take-off on the "baby bell"
breakup decade earlier).
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Re: The Weird OS Built Around a Database
Newsgroups: alt.folklore.computers
Date: Wed, 29 Oct 2025 14:01:09 -1000
rbowman <bowman@montana.com> writes:
I've never used that, only DB2. I have used Raima's db_Vista. It was a C
API and fast compared to big RDMS databases. For its day it was quite
sophisticated. It arguably would have been better than Esri's choice of
dBase for shapefiles.
when I transfered to SJR on west coast, I worked with Jim Gray and
Vera Watson on original SQL/relational, System/R ... and helped with
transfer technology to Endicott ("under the radar" while company was
involved with the next great DBMS "EAGLE" ... to replace IMS) for
SQL/DS. Then when "EAGLE" implodes there is request for how fast can
System/R be ported to MVS, which is eventually released as "DB2"
(originally for decision/support only) ... Jim refs:
https://jimgray.azurewebsites.net/https://en.wikipedia.org/wiki/Jim_Gray_(computer_scientist)
and Vera Watson
https://en.wikipedia.org/wiki/Vera_Watson
... System/R
https://en.wikipedia.org/wiki/IBM_System_R
System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM S/88
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM S/88
Date: 30 Oct, 2025
Blog: Facebook
Some of the MIT CTSS/7094
https://en.wikipedia.org/wiki/Compatible_Time-Sharing_System
people went to the 5th flr to do Multics,
https://en.wikipedia.org/wiki/Multicshttps://en.wikipedia.org/wiki/Multics-like
which spawned both Unix
https://en.wikipedia.org/wiki/Unix
and Stratus (S/88)
https://en.wikipedia.org/wiki/Stratus_Technologieshttps://en.wikipedia.org/wiki/Stratus_VOS
... and others went to the 4th flr to the IBM Cambridge Science Center
and did virtual machines (1st wanted 360/50 to modify with virtual
memory but all the extras were going to FAA/ATC, so had to settle for
360/40 and did CP/40 which morphs into CP/67 when 360/67 standard with
virtual memory becomes available), science center wide-area network
(that morphs into the corporate internal network, larger than
arpanet/internet from just about the beginning until sometime mid/late
80s about the time it was forced to convert to SNA/VTAM; also used for
the corporate sponsored univ BITNET), bunch of other stuff. Had some
friendly rivalry with the 5th flr.
When decision was made to add virtual memory to all 370s, some of the
CSC people move to the 3rd flr, taking over the IBM Boston Programming
System for the VM/370 development group.
1988, Nick Donofrio approved HA/6000, originally for NYTimes to move
their newspaper system (ATEX) off DEC VAXCluster to RS/6000. I rename
it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix).
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester
and mainframe/POK complain they couldn't meet requirements). I had
coined disaster survivability and geographic survivability
(as counter to disaster/recovery) when out marketing HA/CMP. One of
the visits to 1-800 bellcore development showed that S/88 would use a
century of downtime in one software upgrade while HA/CMP had a couple
extra "nines" (compared to S/88) One of the visits to 1-800 bellcore
development showed that S/88 would use a century of downtime in one
software upgrade while HA/CMP had a couple extra "nines" (compared to
S/88)
Also previously worked on original SQL/relational, System/R with Jim
Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
other trivia: Kildall worked on (virtual machine) IBM CP/67 at npg
https://en.wikipedia.org/wiki/Naval_Postgraduate_School
before developing CP/M (name take-off on CP/67).
https://en.wikipedia.org/wiki/CP/M
which spawns Seattle Computer Products
https://en.wikipedia.org/wiki/Seattle_Computer_Products
which spawns MS/DOS
https://en.wikipedia.org/wiki/MS-DOS
more trivia, original SQL/Relational
https://en.wikipedia.org/wiki/IBM_System_R
had been developed on VM/370. Technology transfer to Endicott for
SQL/DS ("under the radar" while company was preoccupied with the next,
great DBMS, "EAGLE"). When "EAGLE" implodes, request for how fast
could System/R be ported to MVS, which is eventually released as DB2,
originally for decision/support only.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67l, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
BITNET (& EARN) posts
https://www.garlic.com/~lynn/subnetwork.html#bitnet
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmpcontinuous availability, disaster survivability, geographic
survivability posts
https://www.garlic.com/~lynn/submain.html#available
posts mentioning assurance
https://www.garlic.com/~lynn/subintegrity.html#assurance
System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Disks
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Disks
Date: 31 Oct, 2025
Blog: Facebook
1981, original 3380, 3mbyte/sec transfer and datastreaming channels
3mbyte/sec (doubled channel data rate, and increased max channel cable
length), each model 2520mb, two sealed disks 1260mb, each disk with
two disk arms that each had 630mb.
http://www.bitsavers.org/pdf/ibm/dasd/3380/http://www.bitsavers.org/pdf/ibm/dasd/3380/GK10-6311-0_3380_Brochure.pdf
Original disk had 20 track spacings between each data track. They then
cut data track spacing in half for double capacity, and then cut track
spacing again for triple capacity. NOTE: already transitioning CKD to
fixed-block (no CKD have been made for decades, all simulating on
industry standard fixed-block devices), can be seen in 3380
records/track formulas where record length has to be rounded up to
multiple of fixed cell size.
Next (1989) were 3390 disks
http://www.bitsavers.org/pdf/ibm/dasd/3390/https://www.extremetech.com/extreme/142911-ibm-3390-the-worlds-largest-and-most-expensive-hard-drive-teardown
When I transferred out to SJR, I got to wander around to IBM (and
non-IBM) silicon valley datacenters, inbluding DASD bldg14
(engineering) and bldg15 (product test) across the street. At the
time, they were running pre-scheduled, 7x24, stand alone testing and
had mentioned that they had recently tried MVS, but it had 15min
mean-time-between failure (in that environment), requiring manual
re-ipl. I offer to rewrite I/O supervisor to make it bullet-proof and
never fail, so they can do any amount of on-demand concurrent testing,
significantly improving productivity. Later, just before 3880/3830
first-customer-ship, FE had a test suite of 57 simulated errors they
thought likely to occur and MVS was failing for all 57 errors
(requiring re-ipl) and in 2/3rds of errors, no indication of what was
causing the failure.
Bldg15 got first engineering 3033 (outside POK processor engineering)
for testing. Because testing only took a percent or two of processor,
we scrounge up 3830 controller and 3330 string and put up private
online service. At the time, air bearing simulation (for thin-film
head design) was only getting a couple turn arounds a month on SJR
370/195. We set it up on bldg15 3033 (slightly less than half 195
MIPS) and they could get several turn arounds a day. Used in 3370FBA
then in 3380CKD
https://www.computerhistory.org/storageengine/thin-film-heads-introduced-for-large-disks/
Mid-1978, bldg15 got engineering 4341 and with little tweaking, could
do 3380 testing. Branch office also heard about it and in Jan1979, con
me into doing 4341 benchmark for national lab that was looking at
getting 70 for compute farm (sort of leading edge of the coming
cluster supercomputing tsunami).
Mid-80s, father of RISC cons me into helping with his idea for "wide"
disk head, parallel read-write 16 closely spaced data tracks (and
follow servo tracks on each side of data track grouping). Problem was
50mbyte/sec data transfer while high-end IBM mainframe channels were
still 3mbyte/sec. Then 1988, branch office asks if I can help LLNL
(national lab) standardize some serial stuff they were working with
which quickly becomes fibre channel standard (including some stuff I
had done in 1980) "FCS", initially 1gbit/sec transfer, full-duplex,
aggregate 200mbyte/sec). Then POK mainframe gets around to shipping
ESCON, initially 10mbyte/sec (when it was already obsolete), later
upgraded to 17mbyte/sec
posts mentioning getting to play disk enginneer in bldg14&15
https://www.garlic.com/~lynn/subtopic.html#disk
posts mentioning DASD, CKD, FBA, multi-track search
https://www.garlic.com/~lynn/submain.html#dasd
posts mentioning FCS &/or FICON
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM VTAM/NCP
Date: 01 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#51 IBM VTAM/NCP
trivia: mid-80s, Communication/sna group was trying to block release
of mainframe tcp/ip support. When they lost, they changed strategy and
said they had corporate ownership of everything that crossed
datacenter walls, it had to ship through them ... what ships got
44kbyte/sec aggregate throughput using nearly whole 3090 CPU. I then
add RFC1044 support and in tuning tests at Cray Research between Cray
and 4341, got sustained 4341 channel throughput using only modest
amount of 4341 processor (something like 500 times improvement in
bytes moved per instruction executed).
Late 80s, a senior disk engineer gets talk scheduled at annual,
internal, world-wide communication group conference, supposedly on
3174 performance. However, his opening was that the communication
group was going to be responsible for the demise of the disk
division. The disk division was seeing drop in disk sales with data
fleeing mainframe datacenters to more distributed computing friendly
platforms. The disk division had come up with a number of solutions,
but they were constantly being vetoed by the communication group (with
their corporate ownership of everything that crossed the datacenter
walls) trying to protect their dumb terminal paradigm. Senior disk
software executive partial countermeasure was investing in distributed
computing startups that would use IBM disks (he would periodically ask
us to drop in on his investments to see if we could offer any
assistance).
The SNA group stranglehold on datacenters wasn't just disks and a
couple years later, IBM has one of the largest losses in the history
of US companies and was being re-organized into the 13 "baby blues" in
preparation for breaking up the company ("baby blues" take-off on
"baby bells" breakup a decade earlier.
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
rfc1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
demise of disk division and communication group stranglehold posts
https://www.garlic.com/~lynn/subnetwork.html#emulation
3-tier, ethernet, tcp/ip, etc posts
https://www.garlic.com/~lynn/subnetwork.html#3tier
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Workstations
Date: 01 Nov, 2025
Blog: Facebook
Early 80s, some people from Stanford approached IBM Palo Alto
Scientific Center to ask if IBM would be interested in picking up a
work stations that they had developed. PASC setup up a review and
invited several operations to listen to Stanford presentation. The net
was that some of the IBM operations were claiming that were doing
something much better than the Stanford workstation ... so Stanford
people went off and formed SUN.
AWD workstation division in Austin was formed to do PC/RT. Previously
it had started with ROMP RISC chip for Displaywriter follow-on; when
that was canceled, they pivoted to the unix workstation market
... getting the company that had done PC/IX for the IBM/PC, to do AIX
for ROMP. PC/RT had 16-bit PC/AT bus ... and developed their own
cards. Then RIOS chipset for RS/6000 that had 32-bit microchannel and
AWD was told they couldn't do their own cards, but had to use the
(that the communication group had heavily performance kneecapped) PS2
cards. Example was the PC/RT 4mbit token-ring card had higher
throughput than the PS2 16mbit token-ring card (joke that PC/RT 4mbit
token-ring server would have higher throughput than RS/6000 16mbit
token-ring server).
1988 Nick Donofrio approved HA/6000, originally for NYTimes to move
their newspaper system (ATEX) off DEC VAXCluster to RS/6000. I rename
it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix; I do a distributed lock
manager/DLM with VAXCluster API and lots of scale-up improvements).
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester and
mainframe/POK complain they couldn't meet requirements). Had coined
disaster survivability and geographic survivability (as counter to
disaster/recovery) when out marketing HA/CMP. One of the visits to
1-800 bellcore development showed that S/88 would use a century of
downtime in one software upgrade while HA/CMP had a couple extra
"nines" (compared to S/88).
Also previously worked on original SQL/relational, System/R with Jim
Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Former executive we had reported to, goes over to head up Somerset/AIM
(Apple, IBM, Motorola), single chip RISC with M88k bus/cache (enabling
clusters of shared memory multiprocessors)
Also 1988, IBM Branch Office asks me if I could help LLNL (national
lab) help standardize some serial stuff they were working with which
quickly becomes fibre-channel stardard, "FCS", including some stuff I
had done in 1980 (initially 1gbit/sec, full-duplex, aggregate
200mbytes/sec) and was planning on using it with commercial cluster
scaleup HA/CMP with 64port non-blocking switch (before kneecaping it
to four system clusters).
trivia: after leaving IBM, was brought in as consultant to a small
client/server startup. Two former Oracle employees (that had been in
the Hester/Ellison meeting) are there responsible for something called
"commerce server" and they want to do payment transactions on the
server. The startup had also invented a technology the called "SSL"
they wanted to use, sometimes called "electronic commerce". I had
responsibility for everything between e-commerce web servers and the
payment networks. I then do a "Why Internet Isn't Business Critical
Dataprocessor" presentation (based on documents, procedures and
software I did for e-commerce) that (Internet RFC/standards editor)
Postel sponsored at ISI/USC.
note, mid90s, i86 chip makers then do hardware layer that translate
i86 instructions into RISC micro-ops for actual execution (largely
negating throughput difference between RISC and i86); 1999 industry
benchmark:
IBM PowerPC 440: 1,000MIPS
Pentium3: 2,054MIPS (twice PowerPC 440)
Sometime after (mainframe) POK announce ESCON, they start some work
with FCS and define a heavy-weight protocol that significantly cut
"FCS" throughput, eventually ships as FICON. IBM publishes 2010 z196
"Peak I/O" benchmark that gets 2M IOPS using 104 FICON (20k
IOPS/FICON). About same time a (native) FCS is announced for E5-2600
server blade claiming over million IOPS (two such FCS having higher
throughput than 104 FICON). Also max-configured z196, goes for $30M
and benchmarked at 50BIPS (625MIPS/core) compared to IBM base list
price for E5-2600 server blade at $1815 and benchmarked at 500BIPS
(and 31BIPS/core). Also IBM pubs recommend that SAPs (system assist
processors that actually do I/O) be kept to 70% CPU (or 1.5M IOPS) and
no new CKD DASD has been made for decades, all being simulated on
industry standard fixed-block devices.
... email (about halfway between when HA/CMP got kneecapped and when
we bailed from IBM) to some people in England:
We recently were on a two week trip thru Europe making calls on
various account teams (and in some cases the account teams had us
present/talk to customers). We went London, Amsterdam, Athens, Nice,
Madrid and back to London. Most recently we were in London from last
weds thru Sunday 3/29 (yesterday). For a more detailed run down talk
to xxxxxx (in NS EMEA marketing) ... they have the names of the
account teams that were at the respective meetings with customers (I
don't remember who the ibm'er was at the northwest water meeting with
David).
... one trip we did presentations at Oracle World, we also did
marketing swings through Asia
801/risc, iliad, romp, rios, pc/rt, rs/6000, power, power/pc posts
https://www.garlic.com/~lynn/subtopic.html#801
original SQL/relational, System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
e-commerce gateway to payment networks
https://www.garlic.com/~lynn/subnetwork.html#gateway
Fibre Channel Standard ("FCS") &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM ACP/TPF
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM ACP/TPF
Date: 01 Nov, 2025
Blog: Facebook
Late 80s, for short stint, my wife was chief architecture for Amadeus
(EU airline res system built off of Eastern "System One" ... 370/195
ACP/TPF), she got replaced by communication group for siding with EU
on using X.25. Didn't do them much good, EU went with X.25 and the
communication group replacement got replaced.
1988 Nick Donofrio approved HA/6000, originally for NYTimes to move
their newspaper system (ATEX) off DEC VAXCluster to RS/6000. I rename
it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix; I do a distributed lock
manager/DLM with VAXCluster API and lots of scale-up improvements).
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester and
mainframe/POK complain they couldn't meet requirements). Had coined
disaster survivability and geographic survivability (as
counter to disaster/recovery) when out marketing HA/CMP. One of the
visits to 1-800 bellcore development showed that S/88 would use a
century of downtime in one software upgrade while HA/CMP had a couple
extra "nines" (compared to S/88).
Also previously worked on original SQL/relational, System/R with Jim
Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
After leaving IBM was brought into the largest ACP/TPF airline res
system to look at the ten impossible things they couldn't do; started
with ROUTES ... got a complete softcopy of OAG (all commercial
schedule airline flts in the world) and redid implementation on Unix
workstation that ran 100 times faster than on ACP/TPF. Then added the
impossible things and only ten times faster ... but after a couple
months was able to demo on unix workstation and show that ten
RS/6000-990s could handle all ROUTE request for all airlines in the
world.
ha/cmp posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
a few posts mentioning Amadeus and routes
https://www.garlic.com/~lynn/2024e.html#92 IBM TPF
https://www.garlic.com/~lynn/2023g.html#90 Has anybody worked on SABRE for American Airlines
https://www.garlic.com/~lynn/2023c.html#8 IBM Downfall
https://www.garlic.com/~lynn/2023.html#96 Mainframe Assembler
https://www.garlic.com/~lynn/2016.html#58 Man Versus System
https://www.garlic.com/~lynn/2015d.html#84 ACP/TPF
--
virtualization experience starting Jan1968, online at home since Mar1970
Tymshare
From: Lynn Wheeler <lynn@garlic.com>
Subject: Tymshare
Date: 02 Nov, 2025
Blog: Facebook
When I transferred out to SJR, I got to wander around IBM (and
non-IBM) datacenters in silicon valley, including Tymshare (and see
them at the monthly BAYBUNCH meetings sponsored by SLAC). Tymshare
started offering their CMS-based online computer conferencing system,
VMSHARE for free in Aug1976 to user group SHARE
https://en.wikipedia.org/wiki/SHARE_(computing)https://www.share.org/
archives here
http://vm.marist.edu/~vmshare/
I cut a deal with Tymshare to get a tape dump of all VMSHARE files for
putting up on internal network and systems ... including online
world-wide sales&marketing support HONE systems (biggest problem I had
was lawyers concerned that internal employees might be contaminated by
being exposed to unfiltered customer information). When M/D bought
Tymshare ... VMSHARE had to be moved to a different platform. I was
also brought into evaluate GNOSIS for the spin-off
https://en.wikipedia.org/wiki/GNOSIShttp://cap-lore.com/CapTheory/upenn/Gnosis/Gnosis.html
also ... Ann Hardy
https://medium.com/chmcore/someone-elses-computer-the-prehistory-of-cloud-computing-bca25645f89
Ann Hardy is a crucial figure in the story of Tymshare and
time-sharing. She began programming in the 1950s, developing software
for the IBM Stretch supercomputer. Frustrated at the lack of
opportunity and pay inequality for women at IBM -- at one point she
discovered she was paid less than half of what the lowest-paid man
reporting to her was paid -- Hardy left to study at the University of
California, Berkeley, and then joined the Lawrence Livermore National
Laboratory in 1962. At the lab, one of her projects involved an early
and surprisingly successful time-sharing operating system.
... snip ...
Ann Hardy at Computer History Museum
https://www.computerhistory.org/collections/catalog/102717167
Ann rose up to become Vice President of the Integrated Systems
Division at Tymshare, from 1976 to 1984, which did online airline
reservations, home banking, and other applications. When Tymshare was
acquired by McDonnell-Douglas in 1984, Ann's position as a female VP
became untenable, and was eased out of the company by being encouraged
to spin out Gnosis, a secure, capabilities-based operating system
developed at Tymshare. Ann founded Key Logic, with funding from Gene
Amdahl, which produced KeyKOS, based on Gnosis, for IBM and Amdahl
mainframes. After closing Key Logic, Ann became a consultant, leading
to her cofounding Agorics with members of Ted Nelson's Xanadu project
... snip ...
On one Tymshare visit, they demo'ed a game that somebody had found on
a Stanford SAIL PDP10 and ported to VM370/CMS, Adventure
https://en.wikipedia.org/wiki/Adventure_game
I got a copy and started making available inside IBM.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
(virtual machine based) online commercial systems
https://www.garlic.com/~lynn/submain.html#online
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
internal corporate network
https://www.garlic.com/~lynn/subnetwork.html#internalnet
dailup-banking posts
https://www.garlic.com/~lynn/submisc.html#dialup-banking
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Doing both software and hardware
Date: 04 Nov, 2025
Blog: Facebook
Did lots of software as undergraduate ... univ hired me responsible
for os/360 ... but also rewrote a lot of (virtual machine) CP/67
... although four of us at the univ, did a clone controller; built 360
channel interface board for Interdata/3 programmed to emulate IBM
terminal/line controller with the addition that it could do auto
speed/baud (I had wanted a single dial-up number/hunt-group for all
terminals and speeds). Later upgraded to Interdata/4 for channel
interface and cluster of Interdata/3s for line/ports (sold by
Interdata and later Perkin-Elmer)
https://en.wikipedia.org/wiki/Interdatahttps://en.wikipedia.org/wiki/Perkin-Elmer#Computer_Systems_Division
after graduating and joining IBM, was sucked into helping with
multithreading 370/195 ... see multhreading ref in this webpage about
death of ACS/360
https://people.computing.clemson.edu/~mark/acs_end.html
195 had out-of-order execution, but no branch prediction (or
speculative execution) and conditional branches drained pipeline so
most code only ran at half rated speed. Adding another i-stream,
simulating 2nd-CPU, each running at half speed, could keep all the
execution units running at full-speed (modulo MVT&MVS 2-CPU
multiprocessor only got 1.2-1.5 the throughput of single CPU; because
of the multiprocessor implementation). Then decision was made to add
virtual memory to all 370s and it was decided that it wasn't partical
to add virtual memory to 370/195 (and any new 195 work, including
multithreading was killed).
IBM had Future System project during 1st half of 70s, was killing off 370 efforts.
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
When FS implodes, there is mad rush to get stuff back into 370 product
pipelines including kicking off quick&dirty 3033&3081 efforts. I got
sucked into helping with tightly-coupled, shared memory 16-CPU 370
system and we con 3033 processor engineers into working on it in their
spare time (lot more interesting that remapping 168 logic to 20%
faster chips). Everybody thought it was great until somebody tells
head of POK (IBM high-end lab) that it could be decades before POK's
favorite son operating system ("MVS") had (effective) tightly-coupled
mutliprocessor support (IBM docs said that MVS 2-CPU systems only had
1.2-1.5 times throughput of 1-CPU system ... because of inefficient
multiprocessor support ... POK doesn't ship a 16-CPU system until
after start of century). Then head of POK invites some of us to never
visit POK again and directs 3033 processor engineers, heads down and
no distractions.
I then transfer out to SJR on the west coast and got to wander around
IBM (and non-IBM) datacenters in silicon valley ... including disk
bldg14 (engineering) and bldg15 (product test) across the street. They
were running prescheduled, 7x24, stand-alone testing. They mentioned
that they had recently tried MVS, but it had 15min MTBF (in that
environment), requiring manual re-IPL. I offer to rewrite I/O
supervisor, making it bullet proof and never fail, allowing any amount
of on-demand, concurrent testing, greatly improving productivity.
Downside was any problems they would point the finger at me and I had
to spend increasing time diagnosing hardware issues and design.
IBM clone controller posts
https://www.garlic.com/~lynn/submain.html#360pcm
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
getting to play disk engineer in bldgs 14&15 posts
https://www.garlic.com/~lynn/subtopic.html#disk
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Mainframe Projects
Date: 06 Nov, 2025
Blog: Facebook
Early 70s, there was Future System effort, completely different and
plan to completely replace 370. I continued to work on 360&370 all
during FS, including periodically ridiculing what they were doing
(which wasn't exactly career enhancing). During FS, 370 efforts were
being killed off (lack of new 370s during FS, claimed to give clone
370 makers their market foothold). When FS
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
implodes, there is mad rush to get stuff back into the 370 product
pipelines, including kicking off quick&dirty 3033&3081 efforts in
parallel. The head of POK was also in the process of convincing
corporate to kill VM370 product, shutdown the development group and
move all the people to POK for MVS/XA. They weren't planning on
telling the people until the very last minute (to minimize the number
that might escape into the Boston/Cambridge area). The information
managed to leak early and several managed to esacpe (DEC VAX/VMS was
in its infancy and joke was that the head of POK was major contributor
to VMS) ... there was hunt for the leaker, but fortunately for me,
nobody gave the them up. Endicott eventually manages to acquire the
VM/370 product mission (for the mid-range), but had to recreate a
development group from scratch
I get con'ed into helping with 125 multiprocessor effort and ECPS for
138/148 (also used for 4331/4341). 115 position memory bus for
microprocessors ... all the same but with different microcoding
including processor for 370. 125 was the same, except the
microprocessor for 370, which was 50% faster. The 125SMP could have up
to five 370 processors (with four left for controller
microprogramming). Old archive post with the ECPS initial analysis for
moving VM370 code into microcode, 6kbytes of VM370 code representing
79.55% of VM370 CPU execution
https://www.garlic.com/~lynn/94.html#21
All the work done for 138/148 ECPS, I was also planning on using for
the 125SMP ... and most of 125SMP support would also be
microcoded. Then Endicott complains that the 125SMP would more than
overlap 148 and managed to get the project killed.
Note: when I joined IBM, one of my hobbies was enhanced production
operating systems for internal datacenters (and online sales&marketing
support HONE systems was one of the 1st and long time customer
originally with CP67 before vm370, also got my 1st overseas business
trips when asked to go along for some of the non-US HONE
installs). Then with the decision to add virtual memory to 370s, also
decided to do VM370 and some of the CSC CP67 people go to the 3rd flr
taking over the Boston Programming Center for VM370. In the morph of
CP67 to VM370 lots of features were simplified or dropped (including
SMP support). Then in 1974, I start adding bunch CP67 stuff into a
VM370R2-base for my CSC/VM (including kernel re-org needed for SMP,
but not SMP itself). Then with a VM370R3-base, I add multiprocessor
support back in, originally for HONE so they can upgrade all their
1-CPU 168s to 2-CPU (US HONE had consolidated all their datacenters in
Palo Alto; trivia when FACEBOOK 1st moves into silicon valley, it is
into a new bldg built next door to the former consolidated US HONE
datacenter, HONE clones were also starting to sprout up all over the
world). BTW, with some optimization the US HONE CSC/VM SMP 2-CPU
systems were getting twice throughput of previous 1-CPU. The 2-CPU,
2-times throughput, further aggravated head of POK who was in the
process of getting VM370 product killed (and MVS was only showing
2-CPU systems with 1.2-1.5 throughput by comparison).
With 125SMP killed, I get asked to help with high-end 16-CPU system
and we con the 3033 processor engineers into helping in their spare
time (a lot more interesting than remapping 168 chip logic to 20%
faster chip). Everybody thought it was great until somebody tells the
head of POK that it could be decades before POK's favorite son
operating system ("MVS") had "effective" 16-CPU support (at the time
MVS docs claimed 2-CPU support only had 1.2-1.5 times throughput of
1-CPU system; aka because of poor & high overhead multiprocessing
support; POK doesn't ship a 16-CPU system until after the turn of the
century, approx 25yrs later). Then the head of POK directs some of us
to never visit POK again, and the 3033 processor engineers, "heads
down, and no distractions".
I transfer out to SJR on the west coast and get to wander around IBM
(and non-IBM) datacenters in silicon valley, including disk bldgs 14
(engineering) and 15 (product test) across the street. They were
running pre-scheduled, 7x24, stand-alone testing and mentioned that
they had recently tried MVS, but MVS had 15min MTBF (requiring manual
re-ipl) in that environment. I offer to rewrite I/O supervisor making
it bullet proof and never fail, allowing any amount of on-demand,
concurrent testing, greatly improving productivity. Downside they
initial target me whenever they had problem and I had to increasingly
play engineer diagnosing (frequently) design problems. I then write an
internal research report about the I/O integrity work and happen to
mention the MVS 15min MTBF, bringing down the wrath of the MVS
organization on my head.
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
CP67l, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
370/125 SMP, tightly-coupled, shared-memory multiprocessor posts
https://www.garlic.com/~lynn/submain.html#bounce
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
--
virtualization experience starting Jan1968, online at home since Mar1970
LSRAD Report
From: Lynn Wheeler <lynn@garlic.com>
Subject: LSRAD Report
Date: 07 Nov, 2025
Blog: Facebook
I had copy of the SHARE
https://en.wikipedia.org/wiki/SHARE_(computing)https://www.share.org/
LSRAD report
http://www.bitsavers.org/pdf/ibm/share/The_LSRAD_Report_Dec79.pdf
scanned it and went to contribute to bitsavers. Problem was it was
published shortly after congress extended the copyright period and had
devil of a time finding somebody at SHARE that would authorize
contribution.
Preface
This is a report of the SHARE Large Systems Requirements for
Application Development (LSRAD) task force. This report proposes an
evolutionary plan for MVS and VM/370 that will lead to simpler, more
efficient and more useable operating systems. The report is intended
to address two audiences: the uses of IBM's large operating systems
and the developers of those systems.
... snip ...
LSRAD published after head of POK had convinced corporate to kill the
VM370 product and shutdown the development group, transferring all the
people to POK for MVS/XA. Endicott did manage to acquire the VM370
product mission (for the mid-range), but was still in the processes of
recreating a development group from scratch.
Acknowledgements
The LSRAD task force would like to thank our respective employers for
the constant support they have given us in the form of resources and
encourgement. We further thank the individuals, both within and
outside SHARE Inc., who reviewed the various drafts of this report. We
would like to acknowledge the contribution of the technical editors,
Ruth Ashman, Jeanine Figur, and Ruth Oldfield, and also of the
clerical assistants, Jane Lovelette and Barbara Simpson
Two computers systems proved invaluable for producing this
report. Draft copies were edited on the Tymshare VM system. The final
report was produced on the IBM Yorktown Heights experimental printer
using the Yorktown Formatting Language under VM/CMS.
... snip ...
Past posts mentioning LSRAD report
https://www.garlic.com/~lynn/2024d.html#31 Future System and S/38
https://www.garlic.com/~lynn/2024c.html#25 Tymshare & Ann Hardy
https://www.garlic.com/~lynn/2024b.html#90 IBM User Group Share
https://www.garlic.com/~lynn/2023g.html#32 Storage Management
https://www.garlic.com/~lynn/2023e.html#20 Copyright Software
https://www.garlic.com/~lynn/2022.html#122 SHARE LSRAD Report
https://www.garlic.com/~lynn/2014j.html#53 Amdahl UTS manual
https://www.garlic.com/~lynn/2011p.html#11 1979 SHARE LSRAD Report
https://www.garlic.com/~lynn/2011p.html#10 1979 SHARE LSRAD Report
https://www.garlic.com/~lynn/2011n.html#62 1979 SHARE LSRAD Report
https://www.garlic.com/~lynn/2011.html#85 Two terrific writers .. are going to write a book
https://www.garlic.com/~lynn/2010q.html#33 IBM S/360 Green Card high quality scan
https://www.garlic.com/~lynn/2010l.html#13 Old EMAIL Index
https://www.garlic.com/~lynn/2009n.html#0 Wanted: SHARE Volume I proceedings
https://www.garlic.com/~lynn/2009.html#70 A New Role for Old Geeks
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Computing Power Consumption
Date: 08 Nov, 2025
Blog: Facebook
New IBM Almaden Research bldg had problem with heat from PC/RT and
then RS/6000 in nearly every office being power cycled ... the air
condition systems weren't made to handle the spike in heat powering on
every machine in the morning (after being off all night) ... it will
have nearly stabilized by end of day, when it started all over again.
past refs:
https://www.garlic.com/~lynn/2024.html#41 RS/6000 Mainframe
https://www.garlic.com/~lynn/2024.html#5 How IBM Stumbled onto RISC
https://www.garlic.com/~lynn/2000f.html#74 Metric System (was: case sensitivity in file names)
Towards the end of last century cloud and cluster supercomputers
started out sharing a lot of similar design/technology. Early on large
cloud could have score of large megadatacenters around the world, each
with half million or more blade servers with enormous automation,
70-80 staff/megadatacenter, (in 2010) each blade server possibly
500BIPS. The cloud operators so optimized their system costs
(assembling their own systems for fraction of the price of brand name
servers) ... that power & cooling were increasingly major
cost. They then started putting heavy pressure on component makers for
1) power drops to zero when idle, but instant on when needed, having
significantly over provisioned to meet "on-demand" online use and 2)
optimize server power consumption when active (introduction of power
consumption ratings in industry benchmarks; threatening to move to
less powerful chips optimized for battery portable use, the less power
hungry chips more than offset requiring more servers (for lower
aggregate power for cloud services).
Sometime last decade, there was industry press that server component
makers were shipping half their product directly to large cloud
operators (that assembled their own servers) and some brand name
server vendors selling off that business. At one point there was also
articles that use a credit card with large cloud operation to spin-up
a supercompter (that ranked in the top 40 in the world) for a few hrs.
cloud megadatacenter posts
https://www.garlic.com/~lynn/submisc.html#megadatacenter
"on-demand" supercomputer
https://www.garlic.com/~lynn/2025.html#16 On-demand Supercomputer
https://www.garlic.com/~lynn/2024c.html#74 Mainframe and Blade Servers
https://www.garlic.com/~lynn/2024c.html#36 Old adage "Nobody ever got fired for buying IBM"
https://www.garlic.com/~lynn/2024b.html#24 HA/CMP
https://www.garlic.com/~lynn/2023.html#100 IBM Introduces 'Vela' Cloud AI Supercomputer Powered by Intel, Nvidia
https://www.garlic.com/~lynn/2021f.html#18 IBM Zcloud - is it just outsourcing ?
https://www.garlic.com/~lynn/2017c.html#6 How do BIG WEBSITES work?
https://www.garlic.com/~lynn/2015g.html#19 Linux Foundation Launches Open Mainframe Project
https://www.garlic.com/~lynn/2013f.html#74 The cloud is killing traditional hardware and software
https://www.garlic.com/~lynn/2013f.html#35 Reports: IBM may sell x86 server business to Lenovo
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Mainframe to PC
Date: 09 Nov, 2025
Blog: Facebook
Some of the MIT CTSS/7094
https://en.wikipedia.org/wiki/Compatible_Time-Sharing_System
people went to the 5th flr to do Multics,
https://en.wikipedia.org/wiki/Multicshttps://en.wikipedia.org/wiki/Multics-like
which spawned both Unix
https://en.wikipedia.org/wiki/Unix
and Stratus (S/88)
https://en.wikipedia.org/wiki/Stratus_Technologieshttps://en.wikipedia.org/wiki/Stratus_VOS
Others went to the IBM Science Center on the 4th flr and did virtual
machines ... wanted 360/50 to modify with virtual memory, but all the
extra 50s were going to FAA/ATC, so had to settle for a 360/40 to add
virtual memory and did cp40/cms, which morphs into cp67/cms when
360/67 standard with virtual memory became available (precursor to
vm370), science center wide-area network which evolves into the
internal network (larger than arpanet/internet from just about the
beginning until sometime mid/late 80s when force to convert to
SNA/VTAM) technology also used for the corporate sponsored univ
BITNET, invented GML 1969 (decade later morphs into ISO standard SGML
and after another decade morphs into HTML at CERN), and bunch of other
stuff.
Kildall worked on (virtual machine) IBM CP/67 at npg
https://en.wikipedia.org/wiki/Naval_Postgraduate_School
before developing CP/M (name take-off on CP/67).
https://en.wikipedia.org/wiki/CP/M
which spawns Seattle Computer Products
https://en.wikipedia.org/wiki/Seattle_Computer_Products
which spawns MS/DOS
https://en.wikipedia.org/wiki/MS-DOS
Early 70s, IBM had Future System effort, totally different from 370
and going to replace 370 (internal politics killing off 370 efforts,
lack of new 370 different FS credited with giving clone 370 makers
their market foothold). I continued to work on 360/370 all during FS,
including periodically ridiculing what they were doing (which wasn't
exactly career enhancing). When FS finally implodes there is made rush
to get stuff back into 370 product pipelines including kicking off
quick&dirty 3033&3081 in parallel. Last nail in the FS coffin was a
study by the IBM Houston Since Center that if 370/195 apps were redone
for FS machine made out of the fastest available hardware technology,
they would have the throughput of 370/145 (about 30 times slowdown).
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
I then got con'ed into helping Endicott with ECPS for 138/148 (redoing
VM370 kernel code into microcode with 10 times speedup). Old archived
post with initial analysis for ECPS:
https://www.garlic.com/~lynn/94.html#21
I transfer to SJR on west coast and get to wander around IBM (&
non-IBM) datacenters in silicon valley, including disk
bldg14(engineering) and bldg15(product test) across the street. They
were doing pre-scheduled, 7x24, stand-alone mainframe testing and
mentioned they had tried MVS but it had 15min MTBF (in that
environment), requiring re-ipl. I offer to rewrite I/O supervisor to
make it bullet proof and never fail, allowing any amount of on-demand,
concurrent testing, greatly improving productivity. Bldg15 gets early
engineering mainframes for I/O testing, getting the 1st engineering
3033 (outside POK processor engineering). Testing took only a couple
percent of processor so we scrounge up 3830 controller and 3330 string
to setup our own online service. Then 1978 get engineering 4341 and
branch office hears about it. They Jan1979, the con me into doing
benchmark for national lab looking at getting 70 for compute farm
(sort of the precursor to the cluster supercomputing and cluster cloud
megadatacenter tsunami).
In the early 80s, I get HSDT, T1 and faster computer links and battles
with communication group (60s, IBM had 2701 that supported T1 links,
70s going to SNA/VTAM and various issues caped links at 56kbits). Also
working with NSF director and was suppose to get $20M to interconnect
the NSF supercomputer centers. Then congress cuts the budget, some
other things happen and eventually release an RFP (in part based on
what we already had running). NSF 28Mar1986 Preliminary Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet
Also in the early 80s, I got permission to give user group talks on
how ECPS was done. After some of the presentations, Amdahl would
corner me with more questions. They had done MACROCODE (370-like
instructions running in microcode mode) to quickly respond a plethora
of 3033 microcode changes required to run MVS. At the time, they were
then in the process of implementing HYPERVISOR ("multiple domain",
virtual machine subset microcode) ... IBM wasn't able to respond with
LPAR/PRSM until nearly decade later. trivia: in the wake of FS
implosion, the head of POK was convincing corporate to kill VM370
product, shutdown the development group and transfer all the people to
POK for MVS/XA (Endicott managed to save the VM370 for the mid-range,
but had to recreate a development group from scratch). Some of the
VM370 people had done a simplified virtual machine for MVS/XA
development, VMTOOL ... it required SIE instruction to move in&out of
virtual machine mode, but 3081 didn't have enough microcode space
... so it was constantly "paging" making it quiet slow. Then customers
weren't converting from MVS to MVS/XA as planned (a lot of 370/XA was
to compensate for MVS shortcomings), somewhat similar to the earlier
problem getting customers to convert from VS2/SVS to VS2/MVS,
mentioned here
http://www.mxg.com/thebuttonman/boney.asp
However, Amdahl was having more success moving customers to MVS/XA
with "Multiple Domain", being able to run MVS & MVS/XA
concurrently. Note 308x was only going to be multiprocessor (no single
processors) and the initial 2-CPU 3081D had less aggregate MIPS than a
single processor Amdahl. IBM responds by doubling processor cache for
3081K, bringing aggregate MIPS up to about the same as Amdahl single
processor (althoug MVS docs had MVS 2-CPU support only getting 1.2-1.5
times the throughput of single processor because enormous MVS 2-CPU
overhead, meaning that MVS 3081K with same aggregate MIPS was still
less throughput than 1-CPU Amdahl). Then some Amdahl people do 370
emulation on SUN which was also ported to PCs, and some early PC
virtual machine.
At the time of the ARPANET cutover to internetworking on 1Jan1983,
there were approx. 100 IMP nodes and 255 hosts. By comparison the
internal network was rapidly approaching 1000 world-wide (not
SNA/VTAM). Old archive post with list of corporate locations that
added one or more network nodes in 1983
https://www.garlic.com/~lynn/2006k.html#8
1988 branch office asks me if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard, "FCS" (not First Customer Ship),
initially 1gbit/sec transfer, full-duplex, aggregate 200mbye/sec
(including some stuff I had done in 1980). Then POK gets some of their
serial stuff released with ES/9000 as ESCON (when it was already
obsolete, initially 10mbytes/sec, later increased to
17mbytes/sec). Then some POK engineers become involved with FCS and
define a heavy-weight protocol that drastically cuts throughput
(eventually released as FICON). 2010, a z196 "Peak I/O" benchmark
released, getting 2M IOPS using 104 FICON (20K IOPS/FICON). About the
same time a FCS is announced for E5-2600 server blade claiming over
million IOPS (two such FCS having higher throughput than 104 FICON
running over FCS). Also IBM pubs recommend that SAPs (system assist
processors that actually do I/O) be kept to 70% CPU (or 1.5M IOPS) and
no new CKD DASD has been made for decades, all being simulated on
industry standard fixed-block devices. Note: 2010 E5-2600 server blade
(16 cores, 31BIPS/core) benchmarked at 500BIPS (ten times max
configured Z196).
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
BITNET posts
https://www.garlic.com/~lynn/subnetwork.html#bitnet
GML, SGML, HTML posts
https://www.garlic.com/~lynn/submain.html#sgml
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
FCS and/or FICON Posts
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
Mainframe to PC
From: Lynn Wheeler <lynn@garlic.com>
Subject: Mainframe to PC
Date: 09 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#67 Mainframe to PC
modified 360/40 with virtual memory and did CP40 (from '82 SEAS
presentation)
https://www.garlic.com/~lynn/cp40seas1982.txt
note atlas paging & virtual memory, Atlas reference (gone 403?, but
lives free at wayback):
https://web.archive.org/web/20121118232455/http://www.ics.uci.edu/~bic/courses/JaverOS/ch8.pdf
from above:
Paging can be credited to the designers of the ATLAS computer, who
employed an associative memory for the address mapping [Kilburn, et
al., 1962]. For the ATLAS computer, |w| = 9 (resulting in 512 words
per page), |p| = 11 (resulting in 2024 pages), and f = 5 (resulting in
32 page frames). Thus a 220-word virtual memory was provided for a
214- word machine. But the original ATLAS operating system employed
paging solely as a means of implementing a large virtual memory;
multiprogramming of user processes was not attempted initially, and
thus no process id's had to be recorded in the associative memory. The
search for a match was performed only on the page number p.
... snip ...
... somewhat analogous to add virtual memory to all 370s; early last
decade was asked to track down that decision and found staff to
executive making decision. Basically MVT storage management was so bad
that region sizes had to be specified four times larger than used, a
typical 1mbyte 370/165 could only run four concurrent regions,
insufficient to keep system busy and justified. Going to 16mbyte
virtual address space, allowed to increase number regions by factor of
4 times (capped at 15 because of 4bit storage protect keys) ... with
little or no paging (somewhat like running MVT in a CP67 16mbyte
virtual machine). I would periodically drop in on Ludlow doing initial
implementation (on 360/67), a little bit of code to create the virtual
memory table and some simple paging (VS2/SVS) ... the biggest task
(similar to CP67) was EXCP/SVC0 was being passed channel programs with
virtual addresses and channels required real addresses ... a copy of
the channel program had to be made substituting real addresses for
virtual (he borrows CCWTRANS from CP67 for merging into EXCP).
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM CEO 1993
Date: 10 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#69 IBM CEO 1993
Example was late 80s, senior disk engineer got a talk scheduled at
annual communication group world-wide internal conference supposedly
on 3174 performance but open the talk with statement that the
communication group was going to be responsible for demise of the disk
division. The disk division was seeing drop in disk sales with data
fleeing mainframe datacenters to more distributed computing friendly
platforms. They had come up with a number of solutions that were
constantly being vetoed by the communication group (with their
corporate responsibility for everything that cross datacenter walls)
trying to preserve their SNA/VTAM paradigm. It wasn't just disks and a
couple years later, the communication group stranglehold on mainframe
datacenters resulted in IBM having one the largest losses in history
of US companies. The disk division senior executive of software
partial countermeasure was investing in distributed computing
startups that would use IBM disks ... and he would periodically ask us
to drop by his investments to see if we could offer any help.
Note: 1988, Nick Donofrio had approved HA/6000 project, originally for
NYTimes to move their newspaper system (ATEX) off DEC VAXCluster to
RS/6000. I rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix). We were also having some
number of distributed computing systems ported to HA/CMP (like LLNL's
"LINCS" and NCAR's "Mesa Archival")
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester
and mainframe/POK complain they couldn't meet requirements). Had
coined disaster survivability and geographic
survivability (as counter to disaster/recovery) when out marketing
HA/CMP. One of the visits to 1-800 bellcore development showed that
S/88 would use a century of downtime in one software upgrade while
HA/CMP had a couple extra "nines" (compared to S/88).
Also previously worked on original SQL/relational, System/R with Jim
Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
industry MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Communication group responsible for demise of disk divsion
https://www.garlic.com/~lynn/subnetwork.html#emulation
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#disk
posts mentioning availability
https://www.garlic.com/~lynn/submain.html#available
posts mentioning assurance
https://www.garlic.com/~lynn/subintegrity.html#assurance
original sql/relational, System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 370 Virtual Memory
Date: 10 Nov, 2025
Blog: Facebook
Early last decade, I was asked to track down decision to add virtual
memory to all 370s; found staff to executive making the
decision. Basically MVT storage management was so bad that region
sizes had to be specified four times larger than used, as result
1mbyte 370/165 only could run four regions concurrently
... insufficient to keep system busy and justified. Going to 16mbyte
virtual address space would allow number regions to be increased by
factor of four times (modulo 4bit storage protect keys capping it at
15) with little or no paging (sort of like running MVT in CP67 16mbyte
virtual machine) ... aka VS2/SVS. Ludlow was doing initial
implementation on 360/67 in POK and I would stop in
periodically. Basically a little bit of code to build the virtual
memory tables and very simple paging code (since they figured that
paging rate would never exceed 5). Biggest piece of code was
EXCP/SVC0, same problem as CP67 ... applications would invoke SVC0
with channel programs with virtual addresses and channel required real
addresses. He borrows CCWTRANS from CP67 to craft into
EXCP/SVC0. Pieces of email exchange in this old archived post
https://www.garlic.com/~lynn/2011d.html#73
I had started posting that systems were getting larger and faster than
disks were getting faster ... and VS2 was working hard to get around
the 15 limit imposed by 4bit storage protect key ... going to a
separate 16mbyte virtual address space for every region ... aka
VS2/MVS. MVS heavily inherited OS/360 pointer passing APIs, they map
an 8mbyte image of the MVS kernel into every 16mbyte address space (so
kernel calls can take the calling API address pointer and access
calling parameters directly, leaving 8mbytes for
applications). However each subsystem were also moved to their private
16mbyte address space, and to pass a pointer to parameter list they
created the 1mbyte common segment area ("CSA") image in every virtual
address space (leaving 7mbytes for applications. Then because common
area requirements were somewhat proportional to number of subsystems
and concurrent applications, common segment area CSA explodes to
common system area (still "CSA") and by 3033 was pushing 5-6 mbytes,
leaving 2-3 mbytes (but threatening to become 8mbytes, leaving zero
for programs ... aka MVS theoretically could have unlimited number of
virtual address spaces, but as the number increased, CSA requirement
would expand to take over everything besides the kernel image). 370/XA
access registers were to address this problem, calls to subsystem
would switch callers address space to secondary and load subsystem
address space as primary ... allowing subsystems to address caller's
virtual address space (in secondary, w/o using CSA). When the
subsystem returns, caller's address space pointer in secondary, would
be moved back to primary. Because it was taking so long to get to
MVS/XA, a subset of access registers were retrofitted to 3033 as
"dual-address space" mode.
Early 80s, I wrote a tome that disk relative system throughput had
declined by order of magnitude since 360 was announced (disks got 3-5
times faster, but systems got 40-50 times faster). Disk division
executive took exception and assigned the division performance group
to refute my claims. After a couple weeks, they basically came back
and said I had slightly understated the problem. They then respun the
analysis for SHARE presentation about how to better configure disks
for improved system throughput (16Aug1984, SHARE 63, B874).
CSC posts (responsible for CP40 & CP67 virtual machines)
https://www.garlic.com/~lynn/subtopic.html#545tech
a few recent posts mentioning CSA, dual-address space, and B874 share
talk
https://www.garlic.com/~lynn/2024d.html#88 Computer Virtual Memory
https://www.garlic.com/~lynn/2024d.html#24 ARM is sort of channeling the IBM 360
https://www.garlic.com/~lynn/2024c.html#55 backward architecture, The Design of Design
https://www.garlic.com/~lynn/2022.html#70 165/168/3033 & 370 virtual memory
https://www.garlic.com/~lynn/2017d.html#61 Paging subsystems in the era of bigass memory
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM 370 Virtual Memory
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 370 Virtual Memory
Date: 11 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#71 IBM 370 Virtual Memory
When "Future System" implodes, there is mad rush to get stuff back
into 370 product pipelines, including kicking off quick&dirty
3033&3081 efforts in parallel. Then the head of POK was convincing
corporate to kill VM370 product, shutdown the VM370 development group
and transfer all the people to POK for MVS/XA. They weren't planning
on telling the people until very last minute (to minimize the number
that might escape into the Boston/Cambridge area). The information
managed to leak and some number managed to escape (DEC VAX/VMS was in
its infancy and joke was that head of POK was major contributor to
VMS). Endicott eventually manages to acquire the VM370 product
mission, but had to recreate a development group from scratch.
With decision to add virtual memory to all 370s, included doing
VM370. Some of the CSC CP67/CMS people split off and move to the 3rd
flr, taking over the IBM Boston Programming Center for the VM370
development group (then as the group exceeds space on the 3rd flr,
they move out to empty IBM SBC bldg at burlington mall (off 128). In
the morph from CP67->VM370 lots of stuff was simplified and/or dropped
(including multiprocessor support).
When I graduated and joined CSC, one of my hobbies was enhanced
production operating systems for internal datacenters ... and the
internal online sales&marketing support HONE systems were one of my
1st (and long time) customers. Then in 1974, I start adding bunch of
stuff back into VM370R2-base for my internal CSC/VM (including kernel
restructure needed for SMP/multiprocessor, but not SMP actual
support). Then move my CSC/VM to VM370R3-base and add SMP back in,
initially for US HONE, so they can upgrade all their 158s&168s to
2-CPU (getting twice the throughput of single processor, at the time
MVS documentation said its 2-CPU support only got 1.2-1.5 times
throughput of 1-CPU).
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67l, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
SMP, tightly-coupled, shared-memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
370/125 5-CPU posts
https://www.garlic.com/~lynn/submain.html#bounce
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM 370 Virtual Memory
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 370 Virtual Memory
Date: 11 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#71 IBM 370 Virtual Memory
https://www.garlic.com/~lynn/2025e.html#72 IBM 370 Virtual Memory
As undergraduate in the 60s at univ, I rewrote large amount of CP67,
including the page replacement algorithm and page I/O. It was global
LRU based (at time when the academic literature was all about "local
LRU" algorithm). After joining IBM, I upgrade my internal CP67 to
global LRU (and also use it for my CSC/VM and SJR/VM internal
systems). In the early 70s, CSC had 768kyte 360/67 (104 pageable 4k
pages) and IBM Grenoble Scientific Center had 1mbyte 360/67 (155
pageable 4k pages) and modified CP67 to correspond to the 60s "Local
LRU" literature ... and both centers ran similar workloads. CSC ran
75-80 users with better throughput and interactive response than the
Grenoble 35 users (with only 104 4k pages compared to grenoble 155
pages)
At Dec81 ACM SIGOPS meeting, Jim Gray asks me if I can help Tandem
co-worker get their Stanford Phd (one of the people that had work on
Stanford ORVYL for 360/67). It involved global LRU page replacements
and the forces from the late 60s ACM local LRU are lobbying to block
giving Phd for anything involving global LRU. I went to send a copy of
the CSC & Grenoble performance data (that could compare the two
strategies on same hardware and some system) but executives block me
from sending for almost a year. I hoped it was punishment for being
blamed for online computer conferencing (and not meddling in academic
dispute).
trivia: In late 70s & early 80s, I was blamed for online computer
conferencing on the internal network (larger than arpanet/internet
from just about the beginning until sometime mid/late 80s, about the
same time it was forced to convert to SNA/VTAM). It really took off
spring of 1981, when I distributed a trip report to visit Jim Gray at
Tandem, only about 300 directly participated, but claims that 25,000
were reading (folklore is that when corporate executive committee was
told, 5of6 wanted to fire me). Some of the results were official
software and officially sanctioned, moderated discussion groups. Also
a researcher was hired to study how I communicated, sat in the back of
my office for 9months, took notes on face-to-face, telephone, got
copies of all incoming&outgoing email (one stat was over 270 different
people/week), logs of instant messages. Results were IBM research
papers, conference talks and papers, books, and Stanford PHD (joint
between language and computer AI).
From IBM Jargon:
Tandem Memos - n. Something constructive but hard to control; a fresh
of breath air (sic). That's another Tandem Memos. A phrase to worry
middle management. It refers to the computer-based conference (widely
distributed in 1981) in which many technical personnel expressed
dissatisfaction with the tools available to them at that time, and
also constructively criticized the way products were [are]
developed. The memos are required reading for anyone with a serious
interest in quality products. If you have not seen the memos, try
reading the November 1981 Datamation summary.
... snip ...
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
paging, replacement algorithms, I/O posts
https://www.garlic.com/~lynn/subtopic.html#clock
online computer conferencing posts
https://www.garlic.com/~lynn/subnetwork.html#cmc
some posts mentioning the paging, local/global LRU, Grenoble, Tandem,
Stanford, computer conferencing
https://www.garlic.com/~lynn/2025c.html#17 IBM System/R
https://www.garlic.com/~lynn/2024g.html#107 IBM 370 Virtual Storage
https://www.garlic.com/~lynn/2024f.html#34 IBM Virtual Memory Global LRU
https://www.garlic.com/~lynn/2024d.html#88 Computer Virtual Memory
https://www.garlic.com/~lynn/2024c.html#94 Virtual Memory Paging
https://www.garlic.com/~lynn/2024b.html#95 Ferranti Atlas and Virtual Memory
https://www.garlic.com/~lynn/2023f.html#109 CSC, HONE, 23Jun69 Unbundling, Future System
https://www.garlic.com/~lynn/2023f.html#25 Ferranti Atlas
https://www.garlic.com/~lynn/2023c.html#90 More Dataprocessing Career
https://www.garlic.com/~lynn/2023c.html#26 Global & Local Page Replacement
https://www.garlic.com/~lynn/2023.html#76 IBM 4341
https://www.garlic.com/~lynn/2022f.html#119 360/67 Virtual Memory
https://www.garlic.com/~lynn/2022d.html#45 MGLRU Revved Once More For Promising Linux Performance Improvements
https://www.garlic.com/~lynn/2018f.html#63 Is LINUX the inheritor of the Earth?
https://www.garlic.com/~lynn/2018f.html#62 LRU ... "global" vs "local"
https://www.garlic.com/~lynn/2014l.html#22 Do we really need 64-bit addresses or is 48-bit enough?
https://www.garlic.com/~lynn/2013k.html#70 What Makes a Tax System Bizarre?
https://www.garlic.com/~lynn/2013i.html#30 By Any Other Name
https://www.garlic.com/~lynn/2012g.html#25 VM370 40yr anniv, CP67 44yr anniv
https://www.garlic.com/~lynn/2012g.html#21 Closure in Disappearance of Computer Scientist
https://www.garlic.com/~lynn/2011c.html#8 The first personal computer (PC)
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM 370 Virtual Memory
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 370 Virtual Memory
Date: 11 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#71 IBM 370 Virtual Memory
https://www.garlic.com/~lynn/2025e.html#72 IBM 370 Virtual Memory
https://www.garlic.com/~lynn/2025e.html#73 IBM 370 Virtual Memory
trivia: before VS2/SVS & VS2/MVS ... Boeing Huntsville had run into
the MVT storage management problem and modified MVTR13 to run in
virtual memory mode on 360/67 (w/o doing any actual paging, but would
reorganize addresses as countermeasure). Lots of places places had
gotten 360/67s for TSS/360 but TSS/360 wasn't coming to production
... so started out running with OS/360 as 360/65. CSC had modified
360/40 with virtual memory and did CP40/CMS ... which morphs into
CP67/CMS when 360/67 started shipping, standard with virtual memory
(and numerous customers then started installing CP67). Univ. of
Michigan (MTS) and Stanford (ORVYL) had done their own virtual memory
systems. Boeing Huntsville configured their 2-CPU system as two two
360/65s with MVT. They ran into the MVT storage management problem
early and modified MVT13 early to run in virtual memory mode.
I had taken two credit hr intro Fortran/computers and at end of
semester I was hired to rewrite 1401 MPIO for 360/30. Univ. was
getting 360/67 for TSS/360, replacing 709/1401 and got 360/30
temporarily (replacing 1401) pending 360/67 shipping. Univ shutdown
datacenter on weekends and I would have the whole place dedicated,
although 48hrs w/o sleep made Monday classes hard. They gave me a pile
of hardware & software manuals and I got to design and implement my
own monitor, device drivers, interrupt handlers, error recovery,
storage management, etc ... and within a few weeks had 2000 card
360/30 assembler program. Within a year of taking intro class, the
360/67 arrives and I'm hired fulltime responsible for OS/360. Student
Fortran job ran under second on 709, but initially over a minute on
360. I install HASP and it cuts the time in half. I then start redoing
STAGE2 SYSGEN, ordering placement of datasets and PDS members to
optimize arm seek and multi-track search, cutting another 2/3rds to
12.9secs. It never got better than 709 until I install UofWaterloo
WATFOR. CSC then comes out and installs CP67 and mostly played with it
during my weekend window, starting to rewrite lots of CP67.
Later, before I graduate, I'm hired fulltime into small group in
Boeing CFO office to help with the formation of Boeing Computer
Service, consolidate all dataprocessing into an independent business
unit. I think Renton datacenter largest in the world, 360/65s arriving
faster than they could be installed, boxes constantly staged in
hallways around machine room (joke that Boeing was getting 360/65s
like other companies got keypunches). Lots of politics between Renton
director and CFO, who only had a 360/30 up at Boeing Field for payroll
(although they enlarge the machine room for a 360/67 for me to play
with when I wasn't doing other stuff). Then the 2-CPU 360/67 is
brought up to Seattle from Boeing Huntsville. Then when I graduate, I
join IBM CSC, rather than staying with Boeing CFO.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
Some recent posts mentioning Univ. student fortran, watfor, Boeing
CFO, renton, huntsville, mvt
https://www.garlic.com/~lynn/2025d.html#99 IBM Fortran
https://www.garlic.com/~lynn/2024f.html#20 IBM 360/30, 360/65, 360/67 Work
https://www.garlic.com/~lynn/2024e.html#136 HASP, JES2, NJE, VNET/RSCS
https://www.garlic.com/~lynn/2024.html#87 IBM 360
https://www.garlic.com/~lynn/2024.html#17 IBM Embraces Virtual Memory -- Finally
https://www.garlic.com/~lynn/2023g.html#19 OS/360 Bloat
https://www.garlic.com/~lynn/2023e.html#34 IBM 360/67
https://www.garlic.com/~lynn/2022e.html#42 WATFOR and CICS were both addressing some of the same OS/360 problems
--
virtualization experience starting Jan1968, online at home since Mar1970
Interactive response
From: Lynn Wheeler <lynn@garlic.com>
Subject: Interactive response
Date: 14 Nov, 2025
Blog: Facebook
One of my hobbies after joining IBM was highly optimized operating
systems for internal datacenters (moving a lot of stuff that I had
done as undergraduate to CP67L for internal systems In early 80s, one
of the 1st (and long time) internal customers, was the internal online
sales&marketing support HONE systems. With the decision to add virtual
memory to all 370s, also decision to do VM370 and some of the CSC
(from 4th flr), take-over the IBM Boston Programming group on the 3rd
flr. In the morph of CP67->VM370, a lot of features were simplified
and/or dropped (like multiprocessor support). In 1974, I started
moving stuff to VM370R2-base for CSC/VM (including kernel reorg for
SMP, but not the actual SMP-support). Then for VM370R3-base CSC/VM, I
put in SMP support, initially for HONE systems, so they could ugprade
their 158&168 systems from 1-CPU to 2-CPU (2-CPU throughput twice that
of 1-CPU, at a time when MVS docs was claiming 2-CPU only had 1.2-1.5
times the throughput of 1-CPU).
I transfer out to SJR, and CSC/VM becomes SJR/VM with further
enhancements was getting .11sec system response for internal
systems. In early 80s, there were increasing studies showing quarter
second response improved productivity. 3272/3277 had .086 hardware
response (plus .11sec system response resulted in .196sec response
seen by users). Then 3274/3278 was introduced with lots of 3278
hardware moved back to 3274 controller, cutting 3278 manufacturing
costs and significantly driving up coax protocol chatter
... increasing hardware response to .3sec to .5sec depending on amount
of data (making it impossible to achieve quarter second). Letters to
the 3278 product administrator complaining about interactive computing
got a response that 3278 wasn't intended for interactive computing but
data entry (sort of electronic keypunch). 3272/3277 required .164sec
system response (for human to see quarter second). I don't believe any
TSO users ever noticed 3278 issues, since they rarely ever saw even
one sec system response).
When I arrived at SJR, it had a MVT 370/195 .... but was shortly
scheduled to get a MVS/168 and a VM/158 (they had a VM/145 that had
been used to develop original SQL/relational RDBMS, System/R, I worked
on it with Jim Gray and Vera Watson). The 168/158 had couple strings
of 3330 with two channel switches connecting to both systems, but
strict orders about MVS packs could never be mounted on 158 3330
strings. One morning operations had mounted a MVS 3330 on VM370
string, and within a few minutes operations were getting irate calls
from all over the bldg about CMS response. One of the issues is MVS
PDS datasets can have multi-cylinder directories that were searched
with multi-track search ... a PDS module load can have a couple
multi-track searches, requiring .317sec for each full cylinder search
(which locks the controller and all associated drives for the duration
of the search). Demands that operations move the offending 3330 was
answered that it would be done 2nd shift. We then have a highly
optimized (for VM) single-pack VS1 mounted on "MVS" drive and start a
PDS directory operation ... and even though the 158 was heavily loaded
... it was able to bring MVS168 nearly to a halt. Operations then
agree to move the "MVS" 3330 if we move VS1 (VS1 had about 1/4 the MVS
pathlength to go from interrupt to redrive of queued channel program)
Later I would be periodically brought into customers multi-CEC MVS
configurations after all the company MVS experts had given up. Example
was 4-CEC 168 shared data configuration ... one of the largest
national grocer, loading store controller apps would almost come to
crawl during peak load. I was brought into class room with large
piles of system activity reports. After 30mins or so, I noticed that
specific 3330 aggregate activity across all systems was peaking at
7/sec (during worst performance). Turns out it was (shared) store
controller PDS dataset ... basically it peaked at aggregate of two
store controller app loads/sec for large hundreds of stores across the
US. Basically for each store controller app load, 1st required avg of
two multi-track searches (one .317secs plus .158secs or .475secs)
Other trivia: when 1st transfer to SJR, I get to wander around IBM
(and non-IBM) datacenters in silicon valley, including disk
bldg14(/engineering) and bldg15(/product test) across the street. They
were doing prescheduled, 7/24, stand alone testing and mentioned that
they had tried MVS, but it had 15min MTBF (in that environment)
requiring manual re-IPL. I offer to rewrite I/O supervisor to make it
bullet-proof and never fail, allowing any amount of on-demand,
concurrent testing, greatly improving productivity. I also got
"channel redrive" pathlength down to about 1/20th that of MVS (and
1/5th of optimized VS1) ... aka from interrupt to redirve of queued
channel program). I then write an internal research report on the I/O
integrity work and happen to mention MVS 15min MTBF ... bringing down
the wrath of the MVS organization on my head.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
DASD, CKD, FBA, multi-track search posts
https://www.garlic.com/~lynn/submain.html#dasd
getting to play disk engineer posts
https://www.garlic.com/~lynn/subtopic.html#disk
some recent posts mentioning interactive response, 3272/3277,
3274/3278
https://www.garlic.com/~lynn/2025e.html#31 IBM 3274/3278
https://www.garlic.com/~lynn/2025c.html#53 IBM 3270 Terminals
https://www.garlic.com/~lynn/2025c.html#47 IBM 3270 Terminals
https://www.garlic.com/~lynn/2025c.html#6 Interactive Response
https://www.garlic.com/~lynn/2025c.html#0 Interactive Response
https://www.garlic.com/~lynn/2025b.html#115 SHARE, MVT, MVS, TSO
https://www.garlic.com/~lynn/2025b.html#47 IBM Datacenters
https://www.garlic.com/~lynn/2025.html#127 3270 Controllers and Terminals
https://www.garlic.com/~lynn/2025.html#69 old pharts, Multics vs Unix
https://www.garlic.com/~lynn/2024f.html#12 3270 Terminals
https://www.garlic.com/~lynn/2024d.html#13 MVS/ISPF Editor
https://www.garlic.com/~lynn/2024c.html#19 IBM Millicode
https://www.garlic.com/~lynn/2024.html#68 IBM 3270
https://www.garlic.com/~lynn/2023g.html#70 MVS/TSO and VM370/CMS Interactive Response
https://www.garlic.com/~lynn/2023f.html#78 Vintage Mainframe PROFS
https://www.garlic.com/~lynn/2023e.html#0 3270
https://www.garlic.com/~lynn/2023b.html#4 IBM 370
https://www.garlic.com/~lynn/2023.html#2 big and little, Can BCD and binary multipliers share circuitry?
https://www.garlic.com/~lynn/2022h.html#96 IBM 3270
https://www.garlic.com/~lynn/2022c.html#68 IBM Mainframe market was Re: Approximate reciprocals
https://www.garlic.com/~lynn/2022b.html#110 IBM 4341 & 3270
https://www.garlic.com/~lynn/2022b.html#33 IBM 3270 Terminals
https://www.garlic.com/~lynn/2022.html#94 VM/370 Interactive Response
https://www.garlic.com/~lynn/2021j.html#74 IBM 3278
https://www.garlic.com/~lynn/2021i.html#69 IBM MYTE
https://www.garlic.com/~lynn/2021c.html#92 IBM SNA/VTAM (& HSDT)
https://www.garlic.com/~lynn/2021.html#84 3272/3277 interactive computing
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM I/O & DASD
Date: 16 Nov, 2025
Blog: Facebook
1980, STL (now SVL) was bursting at the seams and they were moving 300
from the IMS group to offsite bldg, with dataprocessing back to the
STL datacenter. They tried "remote 3270" but found the human factors
unacceptable. I get con'ed into channel extender support so they can
place channel-attached 3270 controllers at the offsite bldg. They
found no perceptible difference in human factors between inside STL
and offsite.
STL had been placing 3270 controllers spread across all channels with
3830 DASD controllers. Then they found the 168s for the offsite IMS
group (with channel extenders) increased throughput by 10-15% (aka
3270 controllers had really high channel busy, but the
channel-extender boxes had significantly lower channel busy for same
amount of 3270 traffic, reducing channel contention with DASD
controllers (and increasing throughput). Then there was some
consideration to put all 3270 controllers on channel extenders (even
those inside STL).
1988, branch office asks if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard ("FCS", including some stuff I had done
in 1980), initially 1gbit/sec transfer, full-duplex, aggregate
200mbyte/sec. Then POK releases some serial stuff with ES/9000 as
ESCON, when it is already obsolete, initially 10mbytes/sec, later
upgraded to 17mbytes/sec.
Then some POK engineers become involved with FCS and define a
heavy-weight protocol for FCS that radically reduces throughput,
eventually released as FICON. Latest public benchmark I've seen is
2010 z196 "Peak I/O" getting 2M IOPS, using 104 FICON (20K
IOPS/FICON). About the same time a FCS is released for E5-2600 server
blades, claiming over a million IOPS (two such FCS having higher
throughput than 104 FICON). Note, IBM docs recommend that SAPs (system
assist processors that do the I/O) be kept to 70% CPU, which would be
1.5M IOPS. Also no CKD DASD has been made for decades, all being
simulated on industry standard fixed-block devices.
trivia: max configured z196 had industry benchmark (number of program
iterations compared to industry reference platform) of 50BIPS (and
$30M). E5-2600 server blade had (same) industry benchmark of 500BIPS
(and IBM list price of $1815). Then industry press had articles that
server component vendors were shipping half their product directly to
large cloud operations (that assemble their own server blades) and IBM
unloads its server blade business.
posts mentioning getting to play disk engineer in bldgs 14&15
https://www.garlic.com/~lynn/subtopic.html#disk
posts mentioning DASD, FBA, CKD, multi-track search
https://www.garlic.com/~lynn/submain.html#dasd
posts mentioning channel extender
https://www.garlic.com/~lynn/submisc.html#channel.extender
posts mentioning FCS and/or FICON
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 8100, SNA, OSI, TCP/IP, Amadeus
Date: 17 Nov, 2025
Blog: Facebook
Evans asked my wife to audit 8100 ... shortly later 8100 was
decommited. Later she did short stint as chief architect of Amadeus
(EU airline system system built off of Eastern "System One" ... 370
ACP/TPF). She sided with EU on x.25 and the ibm communication group
had her replaced. It didn't do them much good, Amadeus went with x.25
anyway and my wife's replacement was replaced.
OSI: The Internet That Wasn't. How TCP/IP eclipsed the Open Systems
Interconnection standards to become the global protocol for computer
networking
https://spectrum.ieee.org/osi-the-internet-that-wasnt
Meanwhile, IBM representatives, led by the company's capable director
of standards, Joseph De Blasi, masterfully steered the discussion,
keeping OSI's development in line with IBM's own business
interests. Computer scientist John Day, who designed protocols for the
ARPANET, was a key member of the U.S. delegation. In his 2008 book
Patterns in Network Architecture(Prentice Hall), Day recalled that IBM
representatives expertly intervened in disputes between delegates
"fighting over who would get a piece of the pie.... IBM played them
like a violin. It was truly magical to watch."
... snip ...
Other trivia: early 80s, got HSDT project, T1 and faster computer
links and battles with communication group (60s, IBM had 2701 that
supported T1 links, 70s going to SNA/VTAM and various issues caped
links at 56kbits). Also working with NSF director and was suppose to
get $20M to interconnect the NSF supercomputer centers. Then congress
cuts the budget, some other things happen and eventually release an
RFP (in part based on what we already had running). NSF 28Mar1986
Preliminary Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet
Communication group was also trying to block release of mainframe
TCP/IP support. When that failed, they changed tactic and said that
since they had corporate responsibility for everything that cross
datacenter walls, it had to be released through them. What shipped got
aggregate 44kbytes/sec using nearly whole 3090 processor. I then added
RFC1044 support and in some tuning tests at Cray Research, between
Cray and 4341, got sustained 4341 channel media throughput using only
modest amount of 4341 processor (something like 500 times improvement
in bytes moved per instruction executed).
Late 80s, my wife was asked to co-author response to gov. agency RFI
... where she included 3-tier network architecture. Then we were were
out making customer executive presentations, highlighting 3-tier,
Ethernet, TCP/IP, Internetworking, high-speed routers, etc (and taking
arrows in the back from the SNA, SAA, & Token-Ring forces)
I was on Chessin's XTP TAB and because there were several
gov. agencies, took XTP to ANSI X3S3.3 as HSP. Eventually X3S3.3 said
that ISO didn't let them standardize anything that didn't conform to
OSI model. XTP/HSP didn't because 1) it supported internetworking
which doesn't exist in OSI, 2) it went directly to LAN MAC bypassing
level 3/4 interface, 3) it went directly to to LAN MAC interface which
doesn't exist in OSI, sitting somewhere in the middle of level
3. There was joke that while IETF (internet) had requirement for two
interoperable implementations before progressing in standards process,
while ISO didn't even require standard be implementable.
1988 Nick Donofrio approved HA/6000, originally for NYTimes to move
their newspaper system (ATEX) off DEC VAXCluster to RS/6000. I rename
it HA/CMP when start doing technical/scientific cluster scale-up with
national labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up
with RDBMS vendors (Oracle, Sybase, Ingres, informix that had DEC
VAXCluster support in same source base as unix; I do a distributed
lock manager/DLM with VAXCluster API and lots of scale-up
improvements).
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester and
mainframe/POK complain they couldn't meet requirements). Had coined
disaster survivability and geographic survivability (as counter to
disaster/recovery) when out marketing HA/CMP. One of the visits to
1-800 bellcore development showed that S/88 would use a century of
downtime in one software upgrade while HA/CMP had a couple extra
"nines" (compared to S/88).
Also previously worked on original SQL/relational, System/R with Jim
Gray and Vera Watson.
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
After leaving IBM, was called in as consultant into small
client/server startup. Two former Oracle employees (that had been in
the Hester/Ellison meeting) are there responsible for something called
"commerce server" and they want to do payment transactions on the
server. The startup had also invented a technology the called "SSL"
they wanted to use, now sometimes called "electronic commerce". I had
responsibility for everything between e-commerce web servers and the
payment networks. I then do a "Why Internet Isn't Business Critical
Dataprocessor" presentation (based on documents, procedures and
software I did for e-commerce) that (IETF Internet RFC/standards
editor) Postel sponsored at ISI/USC.
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
Internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
xtp/hsp posts
https://www.garlic.com/~lynn/subnetwork.html#xtphsp
3tier posts
https://www.garlic.com/~lynn/subnetwork.html#3tier
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmpcontinuous availability, disaster survivability, geographic
survivability posts
https://www.garlic.com/~lynn/submain.html#available
ecommerce payment gateway posts
https://www.garlic.com/~lynn/subnetwork.html#gateway
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: HONE, IBM Unbundling
Date: 20 Nov, 2025
Blog: Facebook
IBM 23jun1969 unbundling announce started to charge for (application)
software (but managed to make case that kernels should still be free),
SE services, maint, etc. SEs training used to involve trainee SEs part
of large group on customer site, but w/unbundling they couldn't figure
out how to NOT charge for trainee SEs onsite. Thus was born HONE,
several HONE virtual machine (CP67, 360/67) datacenters with SE online
access from branches ... "hands-on" experience with guest operating
systems running in virtual machines. Besides doing CP67, CSC had also
ported APL\360 to CP67/CMS as CMS\APL (re-did from swapped 16k
workspaces to large demand paged virtual memory workspaces, having to
redo APL storage management ... and also adding API for system
services, like file I/O, enabling lots of real-world applications) and
HONE starts offering CMS\APL-based sales&marketing support
applications. Eventually the sales&marketing apps came to dominate all
HONE activity (and guest operating system practice just withered
away).
One of my hobbies after graduating and joining CSC was enhanced
production operating systems for internal datacenters and HONE was one
of the 1st (and longtime) customers. Then HONE datacenters were being
spread around the world and I was asked to do the first couple. With
decision to add virtual memory to all 370s, also including doing
VM370. Some of the CSC people split off taking over the IBM Boston
Programming Center on the 3rd flr for the VM370 development group. In
the morph of CP67->VM370, lots of features were simplified and/or
dropped. In 1974, I start my CSC/VM, adding lots of stuff back into a
VM370R2-base (including kernel reorg for multiprocessor support, but
not the actual SMP operation). Then I start into adding actual SMP
support into my VM370R3-base CSC/VM, initially for US HONE so they can
upgrade all their 158&168 systems to 2-CPU. By this time world-wide
HONE systems were easily the largest APL use in the world.
Trivia when facebook 1st moves into silicon valley, it was into a new
bldg built next door to the former consolidated US HONE datacenter.
IBM 23jun1969 unbundling announce posts
https://www.garlic.com/~lynn/submain.html#unbundle
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM Mainframe TCP/IP
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Mainframe TCP/IP
Date: 22 Nov, 2025
Blog: Facebook
Early 1980s, I got HSDT; T1 and faster computer links (both
terrestrial and satellite) and battles with SNA group (60s, IBM had
2701 supporting T1, 70s with SNA/VTAM and issues, links were caped at
56kbit ... and I had to mostly resort to non-IBM hardware). Also was
working with NSF director and was suppose to get $20M to interconnect
the NSF Supercomputer centers. Then congress cuts the budget, some
other things happened and eventually there was RFP released (in part
based on what we already had running). NSF 28Mar1986 Preliminary
Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet. Note RFP had called for T1 links,
however winning bid put in 440kbit/sec links ... then to make it look
something like T1, they put in T1 trunks with telco multiplexors
running multiple 440kbit/sec links over T1 trunks.
Mid-80s, the communication group was fighting off release of mainframe
TCP/IP support (done in pascal/vs). When they lost, they changed
tactics and since they had corporate responsibility for everything
that crossed datacenter walls, it had to be release through them. What
shipped got aggregate 44kbytes/sec using nearly whole 3090
processor. I then add RFC1044 support and in some tuning tests at Cray
Research, between Cray and 4341, got sustained 4341 channel
throughput, using only modest amount of 4341 processor (something like
500 times increase in bytes moved per instruction executed).
The Internet's 100 Oldest Dot-Com Domains
https://www.pcworld.com/article/532545/oldest_domains.html
Co-worker at CSC, responsible for the science centers wide-area
network in the 60s
https://en.wikipedia.org/wiki/Edson_Hendricks
... evolves into internal corporate network (larger than
arpanet/internet from just about the beginning until sometime mid/late
80s, about the time it was forced to convert to SNA/VTAM) and
technology also used for the corporate sponsored univ. BITNET
https://en.wikipedia.org/wiki/BITNET
Account by one of the inventors of GML in 1969 at the science center
about original job was promoting the wide-area network
https://web.archive.org/web/20230402212558/http://www.sgmlsource.com/history/jasis.htm
Actually, the law office application was the original motivation for
the project, something I was allowed to do part-time because of my
knowledge of the user requirements. My real job was to encourage the
staffs of the various scientific centers to make use of the
CP-67-based Wide Area Network that was centered in Cambridge.
... snip ...
I had unix workstation at Interop88 in (non-IBM) booth immediate right
angles to SUN booth, Case was in the SUN booth with SNMP and con'ed
him into installing it on my workstation. trivia: Sunday before the
show opens, the floor nets were crashing with packet floods
.... eventually got diagnosed ... provision about it shows up in
RFC1122.
recent past posts in public group
https://www.garlic.com/~lynn/2024b.html#34 Internet
https://www.garlic.com/~lynn/2024b.html#35 Internet
https://www.garlic.com/~lynn/2024b.html#36 Internet
https://www.garlic.com/~lynn/2024b.html#37 Internet
https://www.garlic.com/~lynn/2024b.html#38 Internet
Also 1988, branch office asks me if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard, "FCS" (not First Customer Ship),
initially 1gbit/sec transfer, full-duplex, aggregate 200mbye/sec
(including some stuff I had done in 1980). Then POK gets some of their
serial stuff released with ES/9000 as ESCON (when it was already
obsolete, initially 10mbytes/sec, later increased to
17mbytes/sec). Then some POK engineers become involved with FCS and
define a heavy-weight protocol that drastically cuts throughput
(eventually released as FICON).
2010, a z196 "Peak I/O" benchmark released, getting 2M IOPS using 104
FICON (20K IOPS/FICON). About the same time a FCS is announced for
E5-2600 server blade claiming over million IOPS (two such FCS having
higher throughput than 104 FICON running over FCS). Also IBM pubs
recommend that SAPs (system assist processors that actually do I/O) be
kept to 70% CPU (or 1.5M IOPS) and no new CKD DASD has been made for
decades, all being simulated on industry standard fixed-block
devices. Note: 2010 E5-2600 server blade (16 cores, 31BIPS/core)
benchmarked at 500BIPS (ten times max configured Z196).
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
GML, SGML, HTML posts
https://www.garlic.com/~lynn/submain.html#sgml
internal network
https://www.garlic.com/~lynn/subnetwork.html#internalnet
BITNET posts
https://www.garlic.com/~lynn/subnetwork.html#bitnet
interop 88 posts
https://www.garlic.com/~lynn/subnetwork.html#interop88
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM 3090
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 3090
Date: 22 Nov, 2025
Blog: Facebook
IBM 3090
https://en.wikipedia.org/wiki/IBM_3090
IBM 3090 page (gone 404, lives on at wayback), Announced Feb 12, 1985
and withdrawn May 5,1989
https://web.archive.org/web/20230719145910/https://www.ibm.com/ibm/history/exhibits/mainframe/mainframe_PP3090.html
In addition, the 3092 processor controller requires two IBM 3370 model
A2 direct access storage devices with string switches or
equivalent. The 3092 requires access to a customer-supplied 3420 model
4, 6 or 8 tape drive or equivalent.
... snip ...
... 3092 service processor started out 4331 running modified version
of VM370R6, before ship was upgraded to pair of 4361s, each with
3370FBA (still running VM370R6) and all service screens were done with
CMS IOS3270.
trivia: During Future System (totally different than 370 and was going
to completely replace it), internal politics was killing off 370
efforts, claims that lack of new 370s during FS gave the clone 370
makers their market foothold. When FS eventually implodes there was
mad rush to get new stuff back into the 370 product pipelines,
including kicking off the quick&dirty 3033&3081 efforts in
parallel. Trivia: one of the last nails in FS coffin was study by IBM
Houston Science Center that showed if 370/195 apps were redone for FS
machine made out of fastest technology available, they would have
throughput of 370/145 (about 30 times slowdown)
https://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.htmlhttp://www.jfsowa.com/computer/memo125.htm
The 370 emulator minus the FS microcode was eventually sold in 1980 as
as the IBM 3081. The ratio of the amount of circuitry in the 3081 to
its performance was significantly worse than other IBM systems of the
time; its price/performance ratio wasn't quite so bad because IBM had
to cut the price to be competitive. The major competition at the time
was from Amdahl Systems -- a company founded by Gene Amdahl, who left
IBM shortly before the FS project began, when his plans for the
Advanced Computer System (ACS) were killed. The Amdahl machine was
indeed superior to the 3081 in price/performance and spectaculary
superior in terms of performance compared to the amount of circuitry.]
... snip ...
After FS implodes, I was also sucked into high-end 16-CPU 370 and we
con the 3033 processor engineers in working on it in their spare
time. Everybody thought 16-CPU was really great until somebody tells
head of POK that it could be decades before POK favorite son operating
system ("MVS") had (effective) 16-CPU support (MVS docs had 2-CPU
support only 1.2-1.5 times throughput of single processor, inefficient
and high overhead SMP support ... POK doesn't ship 16-CPU machine
until after turn of century). Then head of POK invites some of us to
never visit POK again ... and directed 3033 processor engineers heads
down and no distractions. Once the 3033 is out the door, the processor
engineers start on trout/3090 (they periodically had me sneak back
into POK).
Originally, 3081 was multiprocessor only and first was 2-CPU 3081D,
aggregate MIPS less than 1-CPU Amdahl machine. Then they double the
processor cache sizes for the 3081K with about same aggregate MIPS as
1-CPU Amdahl (although MVS 2-CPU 3081K throughput only about .6-.75
times Amdahl machine because of the MVS multiprocessor overhead and
inefficiency, even with about same aggregate MIPS).
1988, branch office asks me if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard, "FCS" (not First Customer Ship),
initially 1gbit/sec transfer, full-duplex, aggregate 200mbye/sec
(including some stuff I had done in 1980). Then POK gets some of their
serial stuff released as ESCON (when it was already obsolete,
initially 10mbytes/sec, later increased to 17mbytes/sec). Then some
POK engineers become involved with FCS and define a heavy-weight
protocol that drastically cuts throughput (eventually released as
FICON). 2010, a z196 "Peak I/O" benchmark released, getting 2M IOPS
using 104 FICON (20K IOPS/FICON). About the same time a FCS is
announced for E5-2600 server blade claiming over million IOPS (two
such FCS having higher throughput than 104 FICON running over
FCS). Also IBM pubs recommend that SAPs (system assist processors that
actually do I/O) be kept to 70% CPU (or 1.5M IOPS) and no new CKD DASD
has been made for decades, all being simulated on industry standard
fixed-block devices. Note: 2010 E5-2600 server blade (16 cores,
31BIPS/core) benchmarked at 500BIPS (ten times max configured Z196).
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM 360, Future System
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 360, Future System
Date: 23 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#78 IBM 360, Future System
Early last decade, I was asked to track down the decision to add
virtual memory to all 370s, I found staff to Evans who had made
decision. Basically MVT storage management was so bad, that regions
had to be specified 4 times larger than used ... as a result, typical
1mbyte 370/165 only ran four regions concurrently, insufficient to
keep system busy and justified. Going to 16mbyte virtual memory (sort
of like running MVT in CP67 16mbyte virtual machine) allowed number of
regions to be increase by factor of four (caped at 15 because of 4bit
storage protect keys) with little or no paging. Ludlow was doing
initial (VS2/SVS) implementation on 360/67 offshift, and I would
periodically drop in on him ... a little bit of work to build virtual
memory tables and some simple paging (expect less than 4-5 page I/Os
per second). Biggest problem was channel programs were now being built
with virtual addresses before calling EXCP/SVC0 and copy channel
programs had to be created replacing virtual addresses with real
(similar to what CP67 did for virtual machines) and Ludlow crafts CP67
"CCWTRANS" into EXCP.
Trivia: At univ. I had taken two credit hr intro to Fortran/Computers
and at end of semester was hired to implement 1401 MPIO in 360
assembler for 360/30. Univ was getting 360/67 for TSS/360 replacing
709/1401 and 360/30 temporarily pending arrival of
360/67. Univ. shutdown datacenter on weekends and I would have it
dedicated (although 48hrs w/o sleep made monday classes hard). I was
given a pile of hardware and software manual and got to design and
implement my own monitor, device drivers, interrupt handlers, error
recovery, storage management, etc ... and within few weeks had 2000
card assembler program. Then within yr of taking into class, 360/67
arrives and I was hired fulltime responsible for OS/360 (tss/360 not
having come to fruition). Then before I graduate, I was hired fulltime
into small group in the Boeing CFO office to help with the formation
of Boeing Computer Services (consolidate all dataprocessing into an
independent business unit). Note Boeing Huntsville had gotten a 2-CPU
360/67 for TSS/360 with lots of 2250s for CAD/CAM work ... but ran it
as two MVT 360/65 systems. They had run into the MVT storage management
problems early and modified MVTR13 to run in (simplified) virtual
memory mode (but w/o paging) as a partial countermeasure (predating
Ludlow's work by quite a few years).
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
a few recent posts mentioning MVT storage management (& Boeing
Huntsville)
https://www.garlic.com/~lynn/2025b.html#95 MVT to VS2/SVS
https://www.garlic.com/~lynn/2025.html#15 Dataprocessing Innovation
https://www.garlic.com/~lynn/2024g.html#106 IBM 370 Virtual Storage
https://www.garlic.com/~lynn/2024e.html#24 Public Facebook Mainframe Group
https://www.garlic.com/~lynn/2023g.html#5 Vintage Future System
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM San Jose and Yorktown Research
Date: 24 Nov, 2025
Blog: Facebook
About the time I (and co-worker) transferred from CSC to SJR in the
70s, I was being blamed for online computer conferencing on the
internal network (co-worker was responsible for the science center
wide-area network that morphs into the internal network and technology
also used for the corporate sponsored univ BITNET) ... which was
larger than arpanet/internet from just about the beginning until
sometime mid/late 80s, about the time it was forced to convert to
SNA/VTAM. It really took off spring of 1981 after I distributed trip
report of visit to Jim Gray at Tandem (had left SJR for Tandem the
fall 1980), only about 300 directly participated, but claims that
25,000 was reading. Folklore is that after the corporate executive
committee was told, 5of6 wanted to fire me. From IBMJargon:
https://web.archive.org/web/20241204163110/https://comlay.net/ibmjarg.pdfTandem Memos - n. Something constructive but hard to control; a fresh
of breath air (sic). That's another Tandem Memos. A phrase to worry
middle management. It refers to the computer-based conference (widely
distributed in 1981) in which many technical personnel expressed
dissatisfaction with the tools available to them at that time, and
also constructively criticized the way products were [are]
developed. The memos are required reading for anyone with a serious
interest in quality products. If you have not seen the memos, try
reading the November 1981 Datamation summary.
... snip ...
After that I was transferred from SJR to YKT ... left in San Jose with
office in SJR (later moved to ALM) and part of wing at LSG ... but had
to commute to YKT a couple times a month.
other triva: I had got HSDT project early 80s, T1 and faster computer
links (both terrestrial and satellite) and battles with SNA group
(60s, IBM had 2701 supporting T1, 70s with SNA/VTAM and issues, links
were caped at 56kbit ... and I had to mostly resort to non-IBM
hardware). Also was working with NSF director and was suppose to get
$20M to interconnect the NSF Supercomputer centers. Then congress cuts
the budget, some other things happened and eventually there was RFP
released (in part based on what we already had running). NSF 28Mar1986
Preliminary Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet. Note RFP had called for T1 links,
however winning bid put in 440kbit/sec links ... then to make it look
something like T1, they put in T1 trunks with telco multiplexors
running multiple 440kbit/sec links over T1 trunks.
Mid-80s, the communication group was fighting off release of mainframe
TCP/IP support (done in pascal/vs). When they lost, they changed
tactics and since they had corporate responsibility for everything
that crossed datacenter walls, it had to be release through them. What
shipped got aggregate 44kbytes/sec using nearly whole 3090
processor. I then add RFC1044 support and in some tuning tests at Cray
Research, between Cray and 4341, got sustained 4341 channel
throughput, using only modest amount of 4341 processor (something like
500 times increase in bytes moved per instruction executed).
CSC (& SJR) co-worker
https://en.wikipedia.org/wiki/Edson_Hendricks
Account by one of the CSC inventors of GML in 1969 at the science
center about original job was promoting the wide-area network
https://web.archive.org/web/20230402212558/http://www.sgmlsource.com/history/jasis.htm
Actually, the law office application was the original motivation for
the project, something I was allowed to do part-time because of my
knowledge of the user requirements. My real job was to encourage the
staffs of the various scientific centers to make use of the
CP-67-based Wide Area Network that was centered in Cambridge.
... snip ...
newspaper article about some of Edson's IBM TCP/IP battles:
https://web.archive.org/web/20000124004147/http://www1.sjmercury.com/svtech/columns/gillmor/docs/dg092499.htm
Also from wayback machine, some additional (IBM missed) references
from Ed's website
https://web.archive.org/web/20000115185349/http://www.edh.net/bungle.htm
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
GML, SGML, HTML posts
https://www.garlic.com/~lynn/submain.html#sgml
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM 3270 Emulators
Date: 27 Nov, 2025
Blog: Facebook
1980, IBM STL (now SVL) was bursting at the seams and they were moving
300 from the IMS group to offsite bldg, with dataprocessing back to
the STL datacenter. They tried "remote 3270" but found the human
factors unacceptable. I get con'ed into channel extender support so
they can place channel-attached 3270 controllers at the offsite
bldg. They found no perceptible difference in human factors between
inside STL and offsite.
STL had been placing 3270 controllers spread across all channels with
3830 DASD controllers. Then they found the 168s for the offsite IMS
group (with channel extenders) increased throughput by 10-15% (aka
3270 controllers had really high channel busy, but the
channel-extender boxes had significantly lower channel busy for same
amount of 3270 traffic, reducing channel contention with DASD
controllers (and increasing throughput). Then there was some
consideration to put all 3270 controllers on channel extenders (even
those inside STL).
1988, branch office asks if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard ("FCS", including some stuff I had done
in 1980), initially 1gbit/sec transfer, full-duplex, aggregate
200mbyte/sec. Then POK releases their serial stuff as ESCON, when it
is already obsolete, initially 10mbytes/sec, later upgraded to
17mbytes/sec.
Then some POK engineers become involved with FCS and define a
heavy-weight protocol for FCS that radically reduces throughput,
eventually released as FICON. Latest public benchmark I've seen is
2010 z196 "Peak I/O" getting 2M IOPS, using 104 FICON (20K
IOPS/FICON). About the same time a FCS is released for E5-2600 server
blades, claiming over a million IOPS (two such FCS having higher
throughput than 104 FICON). Note, IBM docs recommend that SAPs (system
assist processors that do the I/O) be kept to 70% CPU, which would be
1.5M IOPS. Also no CKD DASD has been made for decades, all being
simulated on industry standard fixed-block devices.
trivia: folklore is that token-ring (over CAT wiring) for 3270
terminals was because large installations were finding that 3270 coax
was starting to exceed bldg load limits.
IBM AWD workstation division had done their own adapter card for PC/RT
(PC/AT bus) including 4mbit token-ring card. Then for RS/6000 with
microchannel, they were told they couldn't do their own cards, but had
to use the (communication group heavily performance kneecapped) PS2
microchannel cards, including 16mbit token-ring card. They found that
the PC/RT 4mbit T/R card had higher throughput than the PS2 16mbit T/R
card. New Almaden Research bldg had been heavily provisioned with IBM
CAT wiring, presumming 16mbit T/R; however they found that 10Mbit
Ethernet (over CAT wiring) had lower latency and higher aggregate
throughput, also $69 10Mbit Ethernet cards had higher card throughput
(around 8-8.5mbits/sec) than $800 PS2 microchannel 16Mbit T/R cards.
mid-80s, the communication group was fighting off release of mainframe
TCP/IP support (done in pascal/vs). When they lost, they changed
tactics and since they had corporate responsibility for everything
that crossed datacenter walls, it had to be release through them. What
shipped got aggregate 44kbytes/sec using nearly whole 3090
processor. I then add RFC1044 support and in some tuning tests at Cray
Research, between Cray and 4341, got sustained 4341 channel
throughput, using only modest amount of 4341 processor (something like
500 times increase in bytes moved per instruction executed).
late 80s, some univ did analysis of SNA&TCP/IP ... IBM mainframe
SNA/VTAM LU6.2 had 160k instruction pathlenth while Unix workstation
equivalent, TCP had 5k instruction pathlength
channel extender posts
https://www.garlic.com/~lynn/submisc.html#channel.extender
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
801/risc, iliad, romp, rios, pc/rt, rs/6000, power, power/pc posts
https://www.garlic.com/~lynn/subtopic.html#801
RFC1044 posts
https://www.garlic.com/~lynn/subnetwork.html#1044
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Re-impl Controllers
Date: 28 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#89 Re-impl Controllers
trivia: testing on engineering 3033 in bldg15 only took percent or two
CPU ... so we scavenged 3830 controller and 3330 string and set up our
own online service (including running 3270 coax under road to my
office in SJR/bldg28) . SJR hadn't yet replaced MVT/195 and air
bearing simulation (part of thin-film head design) was getting only a
couple turn arounds a month ... so we sent them up on the bldg15/3033
and they could get several turn arounds a day.
https://en.wikipedia.org/wiki/Thin-film_head#Thin-film_headshttps://www.computerhistory.org/storageengine/thin-film-heads-introduced-for-large-disks/
... 1st used in 3370FBA.
all disks were in process of migrating to fixed-block (can even be
seen in 3380CKD, where records/track formulas had record lengths being
rounded up to multiple of fixed cell size).
more trivia: in the decision to add virtual memory to all 370s
... there was decision to do VM370, in the morph of CP67->VM370, lots
of stuff was simplified and/or dropped. In I/O supervisor rewrite and
I/O integrity work, I also radically cut the pathlength (to about
1/20th that of MVS) ... also added CP67 "CHFREE" back in (an even
faster fastpath after I/O interrupt that as soon as absolutely
possible, would restart channel with queued request).
I had been making references that disk relative system throughput was
approaching decline of order of magnitude (since 360 announce,
i.e. disks were getting 3-5 times faster, but systems were getting
40-50 faster) ... as a result "channel redrive latency" was becoming
increasingly performance issue (lot of 370/xa architecture was trying
to compensate for MVS shortcomings).
posts mentioning getting to play disk engineer in bldgs 14&15
https://www.garlic.com/~lynn/subtopic.html#disk
DASD, CKD, FBA, and multi-track search posts
https://www.garlic.com/~lynn/submain.html#dasd
1980, IBM STL (now SVL) was bursting at the seams and they were moving
300 from the IMS group to offsite bldg, with dataprocessing back to
the STL datacenter. They tried "remote 3270" but found the human
factors unacceptable. I get con'ed into channel extender support so
they can place channel-attached 3270 controllers at the offsite
bldg. They found no perceptible difference in human factors between
inside STL and offsite.
STL had been placing 3270 controllers spread across all channels with
3830 DASD controllers. Then they found the 168s for the offsite IMS
group (with channel extenders) increased throughput by 10-15% (aka
3270 controllers had really high channel busy, but the
channel-extender boxes had significantly lower channel busy for same
amount of 3270 traffic, reducing channel contention with DASD
controllers (and increasing throughput). Then there was some
consideration to put all 3270 controllers on channel extenders (even
those inside STL).
channel-extender posts
https://www.garlic.com/~lynn/submisc.html#channel.extender
some recent posts mentioning disk relative system throughput decline
by order of magnitude
https://www.garlic.com/~lynn/2025e.html#71 IBM 370 Virtual Memory
https://www.garlic.com/~lynn/2025e.html#46 IBM 360/85
https://www.garlic.com/~lynn/2025d.html#108 SASOS and virtually tagged caches
https://www.garlic.com/~lynn/2025d.html#94 IBM VM370 And Pascal
https://www.garlic.com/~lynn/2025d.html#76 Virtual Memory
https://www.garlic.com/~lynn/2025d.html#23 370 Virtual Memory
https://www.garlic.com/~lynn/2025c.html#108 IBM OS/360
https://www.garlic.com/~lynn/2025c.html#79 IBM System/360
https://www.garlic.com/~lynn/2025c.html#20 Is Parallel Programming Hard, And, If So, What Can You Do About It?
https://www.garlic.com/~lynn/2025b.html#33 3081, 370/XA, MVS/XA
https://www.garlic.com/~lynn/2025.html#120 Microcode and Virtual Machine
https://www.garlic.com/~lynn/2025.html#107 IBM 370 Virtual Storage
--
virtualization experience starting Jan1968, online at home since Mar1970
Re-impl Controllers
From: Lynn Wheeler <lynn@garlic.com>
Subject: Re-impl Controllers
Date: 28 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#89 Re-impl Controllers
https://www.garlic.com/~lynn/2025e.html#90 Re-impl Controllers
One of my hobbies after joining IBM was enhanced production systems
for internal datacenter. One of my 1st (and long time) customers were
the internal online sales&marketting support HONE system (initially my
CP67L). In the morph of CP67->VM370, lots of stuff was simplified
and/or dropped. 1974, I start adding a bunch of stuff back into
VM370R2 (including kernel re-org for multiprocessor support, but not
actual SMP, for my CSC/VM. Then for for VM370R3 base, I add more stuff
back in, including SMP support, initially for US HONE so they can
upgrade their 158s & 168s from 1-CPU to 2-CPU.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
In early 80s, there were increasing studies showing quarter second
response improved productivity. 3272/3277 had .086 hardware response
(plus .11sec system response resulted in .196sec response seen by
users). Then 3274/3278 was introduced with lots of 3278 hardware moved
back to 3274 controller, cutting 3278 manufacturing costs and
significantly driving up coax protocol chatter ... increasing hardware
response to .3sec to .5sec depending on amount of data (making it
impossible to achieve quarter second). Letters to the 3278 product
administrator complaining about interactive computing got a response
that 3278 wasn't intended for interactive computing but data entry
(sort of electronic keypunch). 3272/3277 required .164sec system
response (for human to see quarter second). I don't believe any TSO
users ever noticed 3278 issues, since they rarely ever saw even one
sec system response).
1988, branch office asks if I could help LLNL (national lab)
standardize some serial stuff they were working with, which quickly
becomes fibre-channel standard ("FCS", including some stuff I had done
in 1980), initially 1gbit/sec transfer, full-duplex, aggregate
200mbyte/sec. Then POK releases some serial stuff with ES/9000 as
ESCON, when it is already obsolete, initially 10mbytes/sec, later
upgraded to 17mbytes/sec.
Then some POK engineers become involved with FCS and define a
heavy-weight protocol for FCS that radically reduces throughput,
eventually released as FICON. Latest public benchmark I've seen is
2010 z196 "Peak I/O" getting 2M IOPS, using 104 FICON (20K
IOPS/FICON). About the same time a FCS is released for E5-2600 server
blades, claiming over a million IOPS (two such FCS having higher
throughput than 104 FICON). Note, IBM docs recommend that SAPs (system
assist processors that do the I/O) be kept to 70% CPU, which would be
1.5M IOPS. Also no CKD DASD has been made for decades, all being
simulated on industry standard fixed-block devices.
FCS &/or FICON posts
https://www.garlic.com/~lynn/submisc.html#ficon
a few recent posts mentioning 3270s & quarter sec response
https://www.garlic.com/~lynn/2025e.html#75 Interactive response
https://www.garlic.com/~lynn/2025c.html#6 Interactive Response
https://www.garlic.com/~lynn/2024d.html#13 MVS/ISPF Editor
https://www.garlic.com/~lynn/2022c.html#68 IBM Mainframe market was Re: Approximate reciprocals
https://www.garlic.com/~lynn/2021j.html#74 IBM 3278
https://www.garlic.com/~lynn/2021i.html#69 IBM MYTE
https://www.garlic.com/~lynn/2016e.html#51 How the internet was invented
https://www.garlic.com/~lynn/2016c.html#8 You count as an old-timer if (was Re: Origin of the phrase "XYZZY")
https://www.garlic.com/~lynn/2015g.html#58 [Poll] Computing favorities
https://www.garlic.com/~lynn/2014g.html#26 Fifty Years of BASIC, the Programming Language That Made Computers Personal
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: Internet
Date: 30 Nov, 2025
Blog: Facebook
OSI: The Internet That Wasn't. How TCP/IP eclipsed the Open Systems
Interconnection standards to become the global protocol for computer
networking
https://spectrum.ieee.org/osi-the-internet-that-wasnt
Meanwhile, IBM representatives, led by the company's capable director
of standards, Joseph De Blasi, masterfully steered the discussion,
keeping OSI's development in line with IBM's own business
interests. Computer scientist John Day, who designed protocols for the
ARPANET, was a key member of the U.S. delegation. In his 2008 book
Patterns in Network Architecture(Prentice Hall), Day recalled that IBM
representatives expertly intervened in disputes between delegates
"fighting over who would get a piece of the pie.... IBM played them
like a violin. It was truly magical to watch."
... snip ...
Other trivia: early 80s, got HSDT project, T1 and faster computer
links and battles with communication group (60s, IBM had 2701 that
supported T1 links, 70s going to SNA/VTAM and various issues caped
links at 56kbits). Also working with NSF director and was suppose to
get $20M to interconnect the NSF supercomputer centers. Then congress
cuts the budget, some other things happen and eventually release an
RFP (in part based on what we already had running). NSF 28Mar1986
Preliminary Announcement:
https://www.garlic.com/~lynn/2002k.html#12
The OASC has initiated three programs: The Supercomputer Centers
Program to provide Supercomputer cycles; the New Technologies Program
to foster new supercomputer software and hardware developments; and
the Networking Program to build a National Supercomputer Access
Network - NSFnet.
... snip ...
IBM internal politics was not allowing us to bid. The NSF director
tried to help by writing the company a letter (3Apr1986, NSF Director
to IBM Chief Scientist and IBM Senior VP and director of Research,
copying IBM CEO) with support from other gov. agencies ... but that
just made the internal politics worse (as did claims that what we
already had operational was at least 5yrs ahead of the winning bid),
as regional networks connect in, NSFnet becomes the NSFNET backbone,
precursor to modern internet
I was on Chessin's XTP TAB and because there were several
gov. agencies, took XTP to ANSI X3S3.3 as HSP. Eventually X3S3.3 said
that ISO didn't let them standardize anything that didn't conform to
OSI model. XTP/HSP didn't because 1) it supported internetworking
which doesn't exist in OSI, 2) it went directly to LAN MAC bypassing
level 3/4 interface, 3) it went directly to LAN MAC interface which
doesn't exist in OSI, sitting somewhere in the middle of level
3. There was joke that while IETF (internet) had requirement for two
interoperable implementations before progressing in standards process,
while ISO didn't even require standard be implementable.
1988 got HA/6000 approved, originally for NYTimes to move their
newspaper system (ATEX) off DEC VAXCluster to RS/6000. I rename it
HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informix that had DEC
VAXCluster support in same source base as unix; I do a distributed
lock manager/DLM with VAXCluster API and lots of scale-up
improvements).
S/88 Product Administrator started taking us around to their customers
and also had me write a section for the corporate continuous
availability document (it gets pulled when both AS400/Rochester and
mainframe/POK complain they couldn't meet requirements). Had coined
disaster survivability and geographic survivability (as counter to
disaster/recovery) when out marketing HA/CMP. One of the visits to
1-800 bellcore development showed that S/88 would use a century of
downtime in one software upgrade while HA/CMP had a couple extra
"nines" (compared to S/88).
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 benchmarks (number of program iterations compared to
MIPS/BIPS reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
footnote: some of the execs blocking our bidding for NSFnet, were also
involved in transfer of cluster scale-up and restricting us to
4-system cluster.
After leaving IBM, was called in as consultant into small
client/server startup. Two former Oracle employees (that had been in
the Hester/Ellison meeting) are there responsible for something called
"commerce server" and they want to do payment transactions on the
server. The startup had also invented a technology they called "SSL"
they wanted to use, now sometimes called "electronic commerce". I had
responsibility for everything between e-commerce web servers and the
payment networks. I then do a "Why Internet Isn't Business Critical
Dataprocessor" presentation (based on documents, procedures and
software I did for e-commerce) that (IETF Internet RFC/standards
editor) Postel sponsored at ISI/USC.
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
NSFNET posts
https://www.garlic.com/~lynn/subnetwork.html#nsfnet
internet posts
https://www.garlic.com/~lynn/subnetwork.html#internet
OSI XTP/HSP posts
https://www.garlic.com/~lynn/subnetwork.html#xtphsp
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
ecommerce payment gateway posts
https://www.garlic.com/~lynn/subnetwork.html#gateway
--
virtualization experience starting Jan1968, online at home since Mar1970
Microcode
From: Lynn Wheeler <lynn@garlic.com>
Subject: Microcode
Date: 30 Nov, 2025
Blog: Facebook
Palo Alto Science Center did APL microcode assist for 370/145
... measurements showed it ran APL at 168 throughput ... part of
problem was lot of large scale APL required more memory than 145.
Then Endicott cons me into helping with VM370 microcode assist (ECPS)
for 138/148. Was told that there was 6kbyte memory. ECPS did native
microcode instructions at approx 1:1 ... and ordinarily implemented
370 instructions with avg. of 10 native instructions, 6kbytes
accounted for 79.55% of VM370 kernel execution and ECPS reduced by 10
times. Old archived post with intial VM370 analysis for ECPS:
https://www.garlic.com/~lynn/94.html#21
Then Endicott wants preinstall VM370 on every 138/148 shipped (sort of
like LPAR/PRSM done in late 80s original for 3090, well over decade
later). However head of POK was in the process of convincing
corporate to kill VM370 (shutdown the development group and transfer
all the people to POK for MVS/XA) and was able to get preinstall
vetoed. Endicott eventually acquired the VM370 product mission but had
to recreate a development group from scratch.
trivia: 23Jun1969 unbundling announce started to charge for
(application) software (made case kernel software still free), SE
services, maint., etc. SE training use to include part of large group
at customer datacenter. With unbundling, they couldn't figure out how
to NOT charge for trainee SEs at the customer. Thus was born trainee
in branch offices, numerous HONE datacenters around the county, where
branch office SEs could logon, running guest operating systems in CP67
virtual machine. One of my hobbies after graduating and joining IBM
CSC was enhanced production operating systems for internal datacenters
and HONE was one of the 1st (and longtime) customers.
CSC had also ported APL\360 to CP67/CMS as CMS\APL and HONE started
doing CMS\APL-based online sales&marketing apps ... which
eventually came to dominate all HONE use (and practice with guest
operating systems in virtual machines withered away).
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
enhanced production CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
23jun1969 unbundling announce
https://www.garlic.com/~lynn/submain.html#unbundle
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
--
virtualization experience starting Jan1968, online at home since Mar1970
Microcode
From: Lynn Wheeler <lynn@garlic.com>
Subject: Microcode
Date: 30 Nov, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#93 Microcode
Other trivia: with decision to add virtual memory to all 370s, there
was also decision to do VM370. In the morph of CP67->VM370, lots of
features were simplified and/or dropped (including multiprocessor
support). In 1974, with VM370R2-base, I start adding a lot of stuff
back in for my internal CSC/VM, including kernel re-org for SMP
support (but not the actual SMP support) that HONE upgrades to . Then
moving to VM370R3-base, I also add multiprocessor support into my
CSC/VM, originally for HONE so they could upgrade all their 1-CPU
158s&168s systems to 2-CPU (getting twice the throughput). trivia:
US HONE datacenters had been consolidated in silicon valley and when
FACEBOOK 1st moves in, it is into a new bldg built next door to the
former consolidated US HONE datacenter.
1988 approved HA/6000, originally for NYTimes to move their newspaper
system (ATEX) off DEC VAXCluster to RS/6000. I rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with
RDBMS vendors (Oracle, Sybase, Ingres, informaix that had DEC
VAXCluster support in same source base unix; I do a distributed lock
manager/DLM with VAXCluster API and lots of scale-up improvements).
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late-jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market. 1993 industry benchmarks (number of program iterations
compared to MIPS/BIPS reference platform, not actual instruction
count):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
Former executive we had reported to, goes over to head up Somerset/AIM
(Apple, IBM, Motorola), single chip RISC with M88k bus/cache (enabling
clusters of shared memory multiprocessors)
note, mid90s, i86 chip makers then do hardware layer that translate
i86 instructions into RISC micro-ops for actual execution (largely
negating throughput difference between RISC and i86); 1999 industry
benchmark (count of program iterations compared to industry MIPS/BIPS
reference platfrom)
IBM PowerPC 440: 1,000MIPS
Pentium3: 2,054MIPS (twice PowerPC 440)
2010 benachmark (count of program iterations compared to industry
MIPS/BIPS reference platform)
max configured z196 benchmarks at 50BIPS
XEON E5-2600 server blades benchmarks at 500BIPS (10times z196)
E5-2600 were standard system in large cloud (which could have a score)
megadatacenters (around the world), each megadatacenter with half
million or more E5-2600 server blades (500*500,000 = 250,000,000BIPS,
250,000TIPS), with enormous automation ... a megadatacenter
run/managed w/staff of 70-80 people.
CSC posts
https://www.garlic.com/~lynn/subtopic.html#545tech
enhanced production CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
(cloud) megadatacenter posts
https://www.garlic.com/~lynn/submisc.html#megadatacenter
--
virtualization experience starting Jan1968, online at home since Mar1970
Internet
From: Lynn Wheeler <lynn@garlic.com>
Subject: Internet
Date: 01 Dec, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#92 Internet
1972, Learson tried (and failed) to block bureaucrats, careerists, and
MBAs from destroying Watson culture/legacy, pg160-163, 30yrs of
management briefings 1958-1988
https://bitsavers.org/pdf/ibm/generalInfo/IBM_Thirty_Years_of_Mangement_Briefings_1958-1988.pdf
First half of 70s, was IBM's Future System; FS was totally different
from 370 and was going to completely replace it. During FS, internal
politics was killing off 370 projects and lack of new 370 is credited
with giving the clone 370 makers, their market foothold.
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
F/S implosion, from 1993 Computer Wars: The Post-IBM World
https://www.amazon.com/Computer-Wars-The-Post-IBM-World/dp/1587981394/
... and perhaps most damaging, the old culture under Watson Snr and Jr
of free and vigorous debate was replaced with *SYNCOPHANCY* and *MAKE
NO WAVES* under Opel and Akers. It's claimed that thereafter, IBM
lived in the shadow of defeat ... But because of the heavy investment
of face by the top management, F/S took years to kill, although its
wrong headedness was obvious from the very outset. "For the first
time, during F/S, outspoken criticism became politically dangerous,"
recalls a former top executive.
... snip ...
Late 80s, a senior disk engineer gets talk scheduled at annual,
internal, world-wide communication group conference, supposedly on
3174 performance. However, his opening was that the communication
group was going to be responsible for the demise of the disk
division. The disk division was seeing drop in disk sales with data
fleeing mainframe datacenters to more distributed computing friendly
platforms. The disk division had come up with a number of solutions,
but they were constantly being vetoed by the communication group (with
their corporate ownership of everything that crossed the datacenter
walls) trying to protect their dumb terminal paradigm. Senior disk
software executive partial countermeasure was investing in
distributed computing startups that would use IBM disks (he would
periodically ask us to drop in on his investments to see if we could
offer any assistance).
A couple years later (and 20yrs after Learson's failure), IBM has one
of the largest losses in the history of US companies. IBM was being
reorganized into the 13 "baby blues" in preparation for breaking up
the company (take off on "baby bells" breakup decade earlier).
https://web.archive.org/web/20101120231857/http://www.time.com/time/magazine/article/0,9171,977353,00.htmlhttps://content.time.com/time/subscriber/article/0,33009,977353-1,00.html
We had already left IBM but get a call from the bowels of Armonk
asking if we could help with the breakup. Before we get started, the
board brings in the former AMEX president as CEO to try and save the
company, who (somewhat) reverses the breakup and uses some of the same
techniques used at RJR (gone 404, but lives on at wayback)
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
Future System posts
https://www.garlic.com/~lynn/submain.html#futuresys
communication group dumb terminal paradigm
https://www.garlic.com/~lynn/subnetwork.html#terminal
IBM downturn/downfall/breakup posts
https://www.garlic.com/~lynn/submisc.html#ibmdownfall
Pension posts
https://www.garlic.com/~lynn/submisc.html#pensions
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM Downfall
Date: 04 Dec, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#97 IBM Downfall
Gerstner had been president of AMEX ... AMEX & KKR had been in
competition for LBO of RJR and KKR "wins", later KKR runs into
trouble with RJR and hires Gerstner to help.
Then IBM has one of the largest losses in the history of US companies
and was being re-orged into the 13 "baby blues" in preparation for
breaking up the company. Then the board hires Gerstner in an attempt
to save IBM. Account of Gerstner using some of the same techniques
used at RJR as part of saving IBM
https://web.archive.org/web/20181019074906/http://www.ibmemployee.com/RetirementHeist.shtml
Also in 1992, AMEX spins off its card & data processing and
transaction outsourcing business in the largest IPO up until that time
(which had reported to Gerstner; then when IBM was having one of the
largest losses in the history of US companies) as First Data. Later
after leaving IBM, I was doing some work for FDC for datacenter that
handled everything for half of all credit cards in the US (accounts,
billing, settlement, plastics, real-time transactions, call centers,
etc). It had more than 40 max configured IBM mainframes (@$30M) none
older than 18months, constant rolling upgrades ... number needed to
perform settlement in the overnight batch window. A 450K statement
Cobol application ran on all platforms. Other outsourcing datacenters
including backend processing for nearly every cable TV company in the
US, majority of merchant POS card terminals in the US, EFTPS for US
Treasury (99% of withholding and tax payment revenue), etc
Jan1999 I was asked to help try and block the coming economic crises
(we failed). Then in Jan2009, I was asked to webize the 1930s
Pecora/Senate hearings with lots of URLs showing analogies between
what happen then and what happened this time (comment that congress
might have appetite for doing something). I worked on it for a couple
months and then get a call saying it won't be needed after all
(something about capital hill had been totally buried under enormous
mountains of wall street money).
LBO & Private Equity posts
https://www.garlic.com/~lynn/submisc.html#private.equity
AMEX president posts
https://www.garlic.com/~lynn/submisc.html#gerstner
pension posts
https://www.garlic.com/~lynn/submisc.html#pension
economic mess posts
https://www.garlic.com/~lynn/submisc.html#economic.mess
regulatory capture posts
https://www.garlic.com/~lynn/submisc.html#regulatory.capture
Percora & glass-steagall posts
https://www.garlic.com/~lynn/submisc.html#Pecora&/orGlass-Steagall
--
virtualization experience starting Jan1968, online at home since Mar1970
From: Lynn Wheeler <lynn@garlic.com>
Subject: SQL
Date: 04 Dec, 2025
Blog: Facebook
transfer out to SJR in '77 and worked with Jim Gray and Vera Watson on
original SQL/relational, System/R implemented on VM/370 370/145
(Codd's office was on flr above) ... then managed to do tech transfer
to Endicott for SQL/DS (while corporation was preoccupied with the
next, great DBMS, EAGLE). Then when EAGLE imploded there was a request
for how fast could System/R be ported to MVS ... which was eventually
released as DB2 (originally for decision support *only*).
https://en.wikipedia.org/wiki/Edgar_F._Coddhttps://en.wikipedia.org/wiki/IBM_System_Rhttps://en.wikipedia.org/wiki/IBM_SQL/DS
has reference to 1990 IBM Toronto lab doing simplified RDMBS in C for
non-mainframe platforms. Also mentions bringing (mainframe) DB2
product to VM370 platform (even though original System/R was done on
VM370 ... sort of bringing it full-circle).
https://en.wikipedia.org/wiki/IBM_Db2
as well as in mid-90s parallel DB2 for AIX.
https://en.wikipedia.org/wiki/IBM_Db2#Db2_for_LUW
However, in 1988 got approval for HA/6000, originally for NYTimes to
port their newspaper system (ATEX) from DEC VAXCluster to RS/6000. I
rename it HA/CMP
https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing
when I start doing technical cluster scaleup with national labs (LANL,
LLNL, NCAR, etc) and commerical cluster scaleup with RDBMS vendors
(Oracle, Ingres, Sybase, Informix) that had VAXCluster support in same
source base with UNIX (I do a distributed lock manager supporting
VAXCluster semantics to help simplify the ports).
Early Jan1992, meeting with Oracle CEO, IBM AWD executive Hester tells
Ellison that we would have 16-system clusters mid92 and 128-system
clusters ye92. Mid-Jan1992, convinced FSD to bid HA/CMP for
gov. supercomputers. Late Jan1992, HA/CMP is transferred for announce
as IBM Supercomputer (for technical/scientific *ONLY*), and we were
told couldn't work on clusters with more than 4-systems, then leave
IBM a few months later.
Some speculation that it would eat the mainframe in the commercial
market (that and IBM's non-mainframe "DB2" was far from having any
clustering support). 1993 benchmarks (number of program iterations
compared to MIPS/BIPS industry reference platform):
ES/9000-982 : 8CPU 408MIPS, 51MIPS/CPU
RS6000/990 : 126MIPS, 16-systems: 2BIPS, 128-systems: 16BIPS
... snip ...
System/R posts
https://www.garlic.com/~lynn/submain.html#systemr
HA/CMP posts
https://www.garlic.com/~lynn/subtopic.html#hacmp
--
virtualization experience starting Jan1968, online at home since Mar1970
370 138/148 ECPS
From: Lynn Wheeler <lynn@garlic.com>
Subject: 370 138/148 ECPS
Date: 05 Dec, 2025
Blog: Facebook
enlarge and enhance, says 148.
Endicott cons me into doing microcode assist ("ECPS") for
138/148. Says there was 6kbytes for microcode ... and I was to
identify the highest used kernel pathlengths accounting for 6kbytes
(79.55% of kernel execution) ... which ran ten times faster than the
kernel 370 instructions. Then they con me into going around the world
with them presenting the business case justification.
Endicott tries to get corporate to allow them to pre-install VM370 on
every machine shipped (sort of like early LPAR) ... but head of POK
was in the process of convincing corporate to kill the VM370 product,
shutdown the development group and transfer all the people to POK for
MVS/XA. Endicott eventually was able to save the VM370 product
mission, but had to recreate a development group from scratch.
part of head of POK was possibly motivated by analysis done by CERN
comparing VM370/CMS and MVS/TSO presented at SHARE ... while
ordinarily SHARE presentations were freely available, copies
circulating inside IBM got stamped "IBM Confidential - Restricted"
(2nd highest classification), available on "need to know" only.
old archived post with initial ECPS analysis
https://www.garlic.com/~lynn/94.html#21
360/370 mcode posts
https://www.garlic.com/~lynn/submain.html#mcode
--
virtualization experience starting Jan1968, online at home since Mar1970
Internet
From: Lynn Wheeler <lynn@garlic.com>
Subject: Internet
Date: 06 Dec, 2025
Blog: Facebook
re:
https://www.garlic.com/~lynn/2025e.html#92 Internet
https://www.garlic.com/~lynn/2025e.html#95 Internet
https://www.garlic.com/~lynn/2025e.html#96 Internet
as said up thread, 3704/3705/3725 caped at 56kbit/sec. while 60s IBM
2701s handled T1. Part of HSDT, I was suppose to try and show some IBM
content and found FSD T1 Zirpel cards for Series/1 ... for
gov. agencies that were finding their 2701 controllers were
failing. Mid-80s, communication group had done executive committee
presentation on why customers wouldn't be interested in T1 before
sometime mid-90s. They showed 3275 "fat-pipe" configurations (parallel
56kbits links operated as single logical link) for 2, 3, ... 5, 6, 7
... showing number of installed parallel links dropped to zero by
seven parallel links. What they didn't know (or didn't want to show)
was typical telco tariff had 6 or 7 56kbit links about the same as a
single T1. Trivial survey we found 200 customer T1 links installed,
customers had just moved to non-IBM hardware and software.
comment to 3705 post a couple months ago, mentions being ask to turn
out as type-1 product, a 3725/NCP/VTAM emulation, a baby bell did on
Series/1 with significant more function, features and significantly
better price-performance, multiple T1 links, etc; also communication
group eventually coming out with 3737 that supported a single T1 link
(but couldn't couldn't keep it running sustained at T1)
... misc. other stuff
https://www.garlic.com/~lynn/2025d.html#47 IBM HSDT and SNA/VTAM
archived email mentioning 3737
https://www.garlic.com/~lynn/2011g.html#email880130https://www.garlic.com/~lynn/2011g.html#email880606https://www.garlic.com/~lynn/2018f.html#email880715https://www.garlic.com/~lynn/2011g.html#email881005
trivia: 60s as undergraduate, I had been hired fulltime responsible
for OS/360. Univ was getting 360/67 for tss/360, replacing 709/1401
... when it arrived, tss/360 wasn't production so ran as 360/65 with
os/360.
Then CSC came out to install (virtual machine) CP/67 (3rd installation
after CSC itself and MIT Lincoln Labs). I then spend a few months
rewriting pathlengths for running OS/360 in virtual machine. Bare
machine test ran 322secs ... initiall 856secs (CP67 CPU
534secs). After a few months I had CP67 CPU down from 534secs to
113secs.
CP67 delivered had 1052 & 2741 support and some automagic that
recognize terminal type and could switch terminal type port scanner
type. Univ had some tty33 & tty35 and I add ascii tty support
integrated with automagic terminal type recognition (trivia: when
ASCII port scanner arrived for the IBM controller, it came in box from
Heathkit). I then want to have a single dial-up number for all
terminal types ("hunt group") ... but IBM had taken short cut and had
hard-wired port line speed.
This kicks off univ project to build our own clone controller,
building 360 channel interface card for Interdata/3 programmed to
emulate IBM controller with addition it could adapt line speed. Then
it was upgraded to Interdata/4 for channel interface and cluster of
Interdate/3s for port interfaces. Interdata and then Perkin-Elmer sell
them as IBM clone controllers (and four of us were written up
responsible for some part of IBM clone controller business)
https://en.wikipedia.org/wiki/Interdatahttps://en.wikipedia.org/wiki/Perkin-Elmer#Computer_Systems_Division
HSDT posts
https://www.garlic.com/~lynn/subnetwork.html#hsdt
CSC posts
https://www.garliic.com/~lynn/subtopic.html#545tech
clone controller posts
https://www.garlic.com/~lynn/submain.html#360pcm
--
virtualization experience starting Jan1968, online at home since Mar1970
IBM CSC, HONE
From: Lynn Wheeler <lynn@garlic.com>
Subject: IBM CSC, HONE
Date: 06 Dec, 2025
Blog: Facebook
Co-worker at cambridge science center was responsible for science
center CP67-based wide-area network ... which morphs into the internal
network (larger than arpanet/internet from just about the beginning
until sometime mid/late 80s ... about the time it was forced into
converting to SNA/VTAM). One of the CSC members responsible for
inventing GML in 1969:
https://web.archive.org/web/20230402212558/http://www.sgmlsource.com/history/jasis.htm
Actually, the law office application was the original motivation for
the project, something I was allowed to do part-time because of my
knowledge of the user requirements. My real job was to encourage the
staffs of the various scientific centers to make use of the
CP-67-based Wide Area Network that was centered in Cambridge.
... snip ...
note: 7094/CTSS RUNOFF had been ported to CP67/CMS as SCRIPT ... then
after GML was invented, GML tag processing was added to CMS
SCRIPT. PROFS group was collecting internal apps for wrapping 3720
menus aground and selected very early version of VMSG for the email
client. When the VMSG author tried to offer the PROFS group a much
enhanced version, they tried to have him separated from the
company. The whole thing quieted down when he demonstrated every PROFS
email had his initials in a non-displayed field. After that, he only
shared the code with me and one other person.
Early last decade, I was asked to track down the decision to add
virtual memory to all 370s. I found a staff member to the executive
making the decision. Basically MVT storage management was so bad that
region sizes had to be specified four times large than used ... as a
result a typical 1mbyte, 370/165 would only run four regions
concurrently, insufficient to keep the system busy and
justified. Going to a 16mbyte virtual address space (sort of like
running MVT in a CP67 16mbyte virtual machine) allowed number
of regions to be increased by factor of four (capped at 15, because of
4bit storage protect keys) with little or no paging.
I would periodically drop in on Ludlow doing initial VS2/SVS
implementation on 360/67 ... a little bit of code building virtual
memory tables and some simple paging (for the infrequent page
faults). The biggest effort was EXCP/SVC0 was now being passed channel
programs with virtual addresses and channels required real addresses
(similar to CP67 running virtual machines). Ludlow than borrows
CCWTRANS (from CP67) for crafting into EXCP to make copies of passed
channel programs, replacing virtual with real addresses.
other trivia: 23Jun1969 unbundling announce, IBM starts charging for
(application) software (was able to make case, kernel software still
free), SE services, maint. etc. SE training including part of large
group at customer location, but couldn't figure out how not to charge
for trainee SEs at customers. Then a number of "HONE" CP67 datacenters
were established for branch office trainee SEs to practice online
operating guest operating systems (running in virtual
machines). The science center had also ported APL\360 to CP67 as
CMS\APL and HONE also started offering online sales&marketing
support CMS\APL-based applications ... which came to dominate all HONE
activity (and guest operating system practice just withered away).
One of my hobbies after graduating and joining IBM CSC was enhanced
production operating systems for internal datacenters and HONE was one
of the 1st (and long time) customers. In the decision to add virtual
memory to all 370s, there was also development group formed to do
CP67/CMS->VM370/CMS ... although a lot of features were simplified
and/or dropped (including multiprocessor support). In 1974, I start
moving a bunch of stuff to VM370R2-base for my initial (internal)
CSC/VM. I then upgrade to VM370R3-base for CSC/VM including adding
back in multiprocessor support (initially for HONE so they could
upgrade all their 370/158 & 370/168 systems to 2-CPU, getting
twice the throughput of 1-CPU systems).
Around this time CERN does a SHARE presentation of their analysis
comparing VM370/CMS and MVS/TSO. SHARE copies were freely available,
except copies inside IBM got stamped "IBM Confidential - Restricted"
(2nd highest security classification), available on need to know only.
Then with the implosion of FS, there is mad rush to get stuff back
into the 370 product pipelines, including kicking off quick&dirty
3033&3081 efforts in parallel:
http://www.jfsowa.com/computer/memo125.htmhttps://en.wikipedia.org/wiki/IBM_Future_Systems_projecthttps://people.computing.clemson.edu/~mark/fs.html
I get asked to help with a 16-CPU 370 effort, ... and we con the 3033
processor engineers into working on it in their spare time (lot more
interesting than remapping 168 logic to 20% faster chips). Everybody
thought it was great until somebody tells head of POK that it could be
decades before POK's favorite son operating system ("MVS") had
(effective) 16-CPU support (aka at the time, MVS documentation claimed
that its 2-CPU support only had 1.2-1.5 times throughput of 1-CPU
systems; note POK doesn't ship 16-CPU machine until after the turn of
the century). Then the head of POK informs some of us to never visit
POK again and directs 3033 processor engineers, heads down and no
distractions.
San Jose area we would get together friday after work and one of the
subjects would how to get the majority of employees to become computer
users. One of the suggestions was do online telephone directory;
however online app had to be faster than somebody doing lookup in
paper copy on desk. We decide that Jim Gray would spend one week doing
CMS super-fast lookup and I would spend one week with applications
that took softcopy of paper telephone books and convert them to online
lookup ... and also start merging email addresses into the online
versions of the paper copy.
One of the problems was lawyers claiming that telephone directories
were IBM Confidential and it would be security violation to start
making them available online.
I assumed several factors contributed to head of POK convincing
corporate to kill the VM370 product, shutdown the development group
and transfer all the people to POK for MVS/XA (Endicott eventually
manages to save the VM370 product mission, but had to recreate a
development group from scratch).
Science center posts
https://www.garlic.com/~lynn/subtopic.html#545tech
internal network posts
https://www.garlic.com/~lynn/subnetwork.html#internalnet
GML, SGML, HTML posts
https://www.garlic.com/~lynn/submain.html#sgml
23jun1969 unbundling announce posts
https://www.garlic.com/~lynn/submain.html#unbundle
HONE posts
https://www.garlic.com/~lynn/subtopic.html#hone
CP67L, CSC/VM, SJR/VM posts
https://www.garlic.com/~lynn/submisc.html#cscvm
SMP, tightly-coupled, shared memory multiprocessor posts
https://www.garlic.com/~lynn/subtopic.html#smp
future system posts
https://www.garlic.com/~lynn/submain.html#futuresys
--
virtualization experience starting Jan1968, online at home since Mar1970