Sujet : Re: What is an N-bit machine?
De : lynn (at) *nospam* garlic.com (Lynn Wheeler)
Groupes : comp.archDate : 30. Nov 2024, 04:42:25
Autres entêtes
Organisation : Wheeler&Wheeler
Message-ID : <871pytpeu6.fsf@localhost>
References : 1 2 3
User-Agent : Gnus/5.13 (Gnus v5.13)
Lawrence D'Oliveiro <
ldo@nz.invalid> writes:
IBM had legendary market power, all the way up to monopoly status.
Whatever it decreed, its market had to follow.
In the wake of the Future System mid-70s implosion, the head of POK
(high-end mainframe) also convinced corporate to kill the (virtual
machine) VM370 product, shutdown the development group and transfer all
the people to POK to work on MVS/XA (i.e. a lot of XA/370 changes were
to address various bloat&kludges in MVS/370). Come 80s, with 3081 and
MVS/XA, customers weren't converting as planned, continuing to run 3081
370-mode with MVS/370. Amdahl was having more success, it had developed
microcode hypervisor/virtual machine ("multiple domain") support and
able to run both MVS/370 and MVS/XA concurrently on the same (Amdahl)
machine (note Endicott did eventually obtain VM370 product
responsibility for the mid-range 370s, but had to recreate a development
group from scratch).
Amdahl had another advantage, initially 3081 was two processor only and
3081D aggregate MIPS was less than the single processor Amdahl
machine. IBM doubles the processor cache sizes for the 2-CPU 3081K,
having about same aggregate MIPs as Amdahl single CPU .... however at
the time, IBM MVS documents had MVS two-processor (multiprocessor
overhead) support only getting 1.2-1.5 times the throughput of single
processor (aka Amdahl single processor getting full MIPS throughput
while MVS two processor 3081K loosing lots of throughput to
multiprocessor overhead).
POK had done a rudementary virtual software system for MVS/XA testing
... which eventually ships as VM/MA (migration aid) and then VM/SF
(system facility) ... however since 370/XA had been primarily focused to
compensate for MVS issues ... 3081 370/XA required a lot of microcode
tweaks when running in virtual machine mode and 3081 didn't have the
space ... so switching in and out of VM/MA or VM/SF with virtual machine
mode, had a lot of overhead "paging" microcode. It was almost a decade
later before IBM was able to respond to Amdahl's
hypervisor/multiple-domain with 3090 LPAR & PR/SM support.
-- virtualization experience starting Jan1968, online at home since Mar1970