[email protected] (Charles Mills) writes:
> independently of z/VM, but that is no longer true.
>
> I'm quite familiar architecturally with CMS. Yes, it is in one sense
> an operating system. If I drew a picture of z/VM with a bunch of
> guests, CMS would be a peer to VSE, z/OS and Linux, all of which are
> no-doubt-about-it operating systems. So yes, it must be an OS. It's a
> funny beast of an OS: a single-user operating system for a mainframe!
> OTOH, the way people use it it is more like a single TSO session than
> an operating system. It is more of a single-user terminal session than
> an OS. And yes, it no longer can be IPLed on the hardware, so a purist
> might say that disqualifies it as an operating system right
> there. What kind of operating system requires an operating system in
> order to run?
>
> Would not the proper phrase be "run independently of CP"? CMS is a
> component of z/VM; it can't be independent of z/VM, it *is* z/VM.

original CMS was "cambridge monitor system" running stand-alone on
360/40. The same 360/40 had virtual memory hardware modifications to
support developement of virtual machine cp/40 in parallel. When standard
360/67 with virtual memory became available, cp/40 morphs into cp/67
... with CMS still running both stand-alone and in virtual machine
... cp40/cms development discussed here
http://www.garlic.com/~lynn/cp40seas1982.txt
also mentioned in
http://www.ibmsystemsmag.com/mainframe/administrator/Virtualization/importance-today/
and my recent posting in this mailing list
http://www.garlic.com/~lynn/2017b.html#30 Virtualization's Past Helps Explain 
Its Current Importance

In the morph from CP67/CMS to VM370/CMS, they dropped and simplified a
lot of CP67, change CMS name to conversational monitor system, and put
test in CMS to cripple it from running on bare hardware.

In the early 80s, CMS under MVS had really bad performance ... not
because of CMS ... but MVS structural issues ...  vm370/cms could
provide .1sec interactive cms response when MVS/CMS was 1sec or worse
with similar workload on same hardware. Some people thought that CMS
might fix the really bad TSO interactive characteristics ... but it
wasn't so much TSO versus CMS ... but underlying MVS structural issues.

In the wake of failure of FS project in the mid-70s (never announced or
shipped, FS would completely replace 360/370 and completely different
and during FS, 370 projects were being shutdown, the lack of 370
products during the FS period is credited with clone makers gaining
market foothold). In the wake of FS failure, there was mad rush to get
370 products back in the pipelines. 303x and 3081 were kicked off in
paralle ... and head of POK managed to convince corporate to kill
vm370/cms product and transfer all the vm370/cms to POK for MVS/XA work
... or otherwise MVS/XA wouldn't be able to meet ship schedule (well
into the 1980s).

Endicott eventually managed to resurrect VM370/CMS product mission
... but had to reconstitute development group from scratch. Aside,
Tymshare had started offering there CMS-based online computer
conferencing to SHARE for free as VMSHARE in Aug1976. In the VMSHARE
archives there are customer comments about code quality during this
period as Endicott rebuilds a vm370/cms development group
http://vm.marist.edu/~vmshare

At the time of the VM370/CMS shutdown there was development of greatly
expanded CMS support for MVS system services, read/write MVS (shared)
disks, etc ... which all disappears with the shutdown .... which was
just fine with POK. After Endicott resurrects VM370, they were much more
interest in CMS having DOS/VS capatibility (than MVS compatibiity; aka
DOS/VS system services).

With 4300s, Endicott manage to start making market position where large
corporations were ordering multiple hundred 4300s at a time for placing
out in departmental areas (sort of leading edge of coming distributed
computing tsunami). This became such a large market, Endicott managed to
get corporate to declare vm370/cms as the strategic interactive
offering.

At the time, the only disks for placing out in non-datacenter
environment were FBA, which MVS didn't support. The requirement was also
for support where it was several tens of systems per support person
... rather than MVS requiring several people per system.

There was also late 70s, where customers were looking at 4341 cluster
having higher throughput than 3033, more storage, more I/O, much lower
environmentals, smaller footprint, and much lower cost. At one point POK
felt so threatened that head of POK managed to get the allocation of
critical 4341 manufacturing component cut in half.

I participated in 4341 benchmarking for LLNL that was looking at getting
70 for a compute farm ... sort of the precursor to the current cluster
supercomputers

Some of the VM370/CMS people that went to POK, did do a virtual machine
implementation to support MVS/XA development ... which was never
intended to be released ... which contributed to idea for CMS under
MVS. During the 80s when customers weren't migrating from MVS to MVS/XA
like they were suppose to ... the internal tool was packaged as the
migration aid to ... aid migration. Then POK had justification for a big
group to bring the migration aid up to the feature, function,
performance of VM370. At the same time, somebody inside IBM had added
full 370-XA support to VM370. There was escalation between POK and
Endicott over VM/XA, whether to have a large development group to bring
migration aid up to vm370 level or to use the VM370 with XA support
already running. For whatever reason, POK won.

When I was undergraduate at the university ... totally redid OS/360
sysgen ... part of old 68fall SHARE presentation:
http://www.garlic.com/~lynn/94.html#18 CP/67 & OS MFT14

Also rewrote a lot of CP67 code.  Original CP67 CPU overhead increased
elapsed running time from 12.9 secs to 34.2secs per job (21.3 secs of
overhead CPU). CP67 pathlength optimization reduced elapsed running time
to 17.4secs/job, CP67 CPU overhead reduced from 21.3sec to 3.9secs (lots
of specific pathlengths reduced by factor of 100 times).

It also mentions doing careful sysgen of OS/360 to reduce standalone
time from over 30secs/job to 12.9secs/job ... careful placements of
datasets and PDS members on disk to optimized arm seek time and
multi-track search time for PDS members. CP67 for "normal" OS would have
looked less bad since instead of nearly tripling elapsed time (12.9+21.3
to 34.2), it would have less than doubled elapsed time (35+21.3 to
56.9). The CP67 pathlength enhancements for standard system would have
been 35+3.9 to 38.9secs ... compared to 12.9+3.9 to 17.4secs (careful
OS/360 sysgen and data ordering significantly reduced wait time to run
the jobs).

Must of CMS CPU time was frequently CP67 doing disk I/O simulation (ala
"CCWTRANS") ... I did a special disk CCW that CMS would use when running
in virtual machine, cut that time by something like 80-90%.  The people
at science center
http://www.garlic.com/~lynn/subtopic.html#545tech

roundly critized me for violating 360 "architecture" ... if I was to do
something that was virtual machine specific, I had to do it with the
"diagnose" instruction ... which is defined as model dependent, and can
have the fiction of "virtual machine" model. Thus was born diagnose I/O
when running in virtual machine (later VM370 crippled CMS being able to
run stand alone).

trivia, recent mention that "CCWTRANS" was also borrowed for moving MVT
to OS/VS2
http://www.garlic.com/~lynn/2017b.html#8 BSAM vs QSAM

-- 
virtualization experience starting Jan1968, online at home since Mar1970

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to