[email protected] (Shmuel Metz  , Seymour J.) writes:
> IBM issued sevaral releases of TSS/360 and even had a PRPQ for TSS/370
> before they cancelled it. TSS performance was considerably better by
> then.

re:
http://www.garlic.com/~lynn/2012o.html#31 Regarding Time Sharing
http://www.garlic.com/~lynn/2012o.html#32 Regarding Time Sharing

1968 ... i did simulated 35 user cp67/cms for fortran program edit,
compile, link and execute ... that had better response and throughput
than simulated four user tss/360 for same fortran program, edit,
compile, link and execute ... on the same exact 360/67 hardware.

at the time of the test, cp67 had many of the pathlength enhancements
mentioned in previous posts ... but not the rewrite of the virtual
memory and my dynamic adaptive resource manager.

tss/370 had number of special bids, one was AT&T to do a stripped down
TSS/370 kernel (SSUP) that had unix layered ontop. Amdahl was in
marketing UTS ... direct unix port ... mostly running in vm370 virtual
machine. 

Eventually IBM also did port of UCLA LOCUS (unix work-alike) ...  also
running in vm370 virtual machine ... announced as aix/370 (coupled with
aix/386 ... LOCUS supported distributed filesystem as well as processing
... even migrating processes between 386 and 370 with different
binaries).

Issue in the period was field maintenanced required EREP and error
recovery. At the time, to add mainframe EREP and error recovery to unix
kernel was several times larger effort than just the straight forward
unix port to mainframe.

one of the early jokes about tss/360 ("official" timesharing) was that
it had approx. 1200 people working on it at time when the science center
had approx 12 people working on cp67/cms. by the mid-70s, after the
tss/360 decommit and before pok convinced corporate to kill-off vm370
and transfer all of the people to POK to work on mvs/xa (mentioned
before that endicott manage to save the product mission, but had to
reconstite a group from scartch) ... the tss/370 group had 20 people and
the vm/370 group (followon to cp67) had maybe 300 people ... aka the
performance of a system is inversely proportional to the number of
people working on it.

spring 1985, i was involved in comparison study of tss and vm/sp
kernels; part of that analysis (vm/sp had gotten grossly bloated by that
time):

             TSS     VM/SP
modules      109     260
lines of 
assembler
code         51k     232k

misc. posts posts mentioning tss/vmsp comparison study:
http://www.garlic.com/~lynn/2001m.html#53 TSS/360
http://www.garlic.com/~lynn/2006e.html#33 MCTS
http://www.garlic.com/~lynn/2010e.html#17 Senior Java Developer vs. MVS Systems 
Programmer (warning: Conley rant)

misc. old email mentioning tss/unix
http://www.garlic.com/~lynn/2006b.html#email800310
http://www.garlic.com/~lynn/2006t.html#email800327
http://www.garlic.com/~lynn/2007e.html#email800404
http://www.garlic.com/~lynn/2006f.html#email800404
http://www.garlic.com/~lynn/2007b.html#email800408
http://www.garlic.com/~lynn/2006e.html#email840109

note: native tss/360 had paged-mapped filesystem ... pagetables could be
setup to point directly at executable image on disk ... and could be
page directly in and executed ... even relocatable code that could
concurrently reside in different virtual address spaces at different
virtual addresses. the tss/360 relocatable executable format was quite a
bit different thatn os/360 relocatble executable format. In os/360
relocable executable format, the file had to be preloaded into memory
and all address constants swizzled to absolute value (representing their
loaded location). tss/360 executable address constants were truely
relocatable ... i.e. the address constants in the executable image
didn't require any fiddling ... the executable image could be mapped
to any virtual address and everything worked.

the os/360 convention caused enormous problems for me when i did page
mapped filesystem for cp67/cms. cms adopted lots of applications,
compilers, loaders, etc. from os/360 ... which required that relocatable
address constants be swizzled to absolute address after loading and
before execution. for truely high-performance operation ... I had to do
a lot of application fiddling to make the relocatable address constants
more like tss/360. I had also seen lots of (other) things that tss/360
had done wrong in the page mapped implementation ... which i avoided
in my cp67/cms implementation. Then Future System effectively did
a similar rote page mapped filesystem desgin with all the other tss/360
performance shortcomings. It was one of the things that I would
periodically ridicule the FS group about ... believing what I
already had running was better than what they were doing high level
specs. on. ... misc. past posts mentioning having done cp67/cms
paged-mapped filesystem
http://www.garlic.com/~lynn/submain.html#mmap
misc. past posts mentioning horrors having to deal with os/360
relocable address constants
http://www.garlic.com/~lynn/submain.html#adcon

-- 
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