[email protected] (Mike Beer) writes:
> https://www-03.ibm.com/systems/z/os/zvse/about/history1970s.html

Endicott told me there was 6kbytes available for assist microcode ...  I
was to identify the highest used code paths in the vm370 kernel for
replication in microcode.
(standard 370 kernel instructions translated on about byte-for-byte
basis)

the low & mid-range 370 native (vertical) microcode emulated 370 on
about 10:1 basis ... so instructions moved from 370 to native code got
approx.  10:1 speedup.

old post with times I did of vm370 kernel for selecting 6k bytes of code
segments for dropping into "ECPS" microcode
http://www.garlic.com/~lynn/94.html#21

6kbyte cutoff accounted for 79.55% of kernel execution ... gets a 10:1
speedup.

At the same time there was VS1 handshaking that bypassed certain VS1
processes and left it to VM370 ... resulting in VS1 under VM370 ran
faster than stand alone on the bare machine.

Endicott then tried to get corporate approval to preinstall vm370 on
every 138&148 shipped from the factory (sort of like LPARs
today). However, this is in the period after Future System implosion and
mad rush to get 370 products back into the IBM product pipeline.  POK
kicked off 3033 & 3081 in parallel and convinced corporate to kill the
vm370 product, shutdown the vm370 development group and move all the
people to POK to work on MVS/XA (or otherwise MVS/XA wouldn't be able to
ship on schedule). Endicott managed to save the VM370 product mission
... but had to reconstitute a development group from scratch ...  but
wasn't able to convince corporate to allow vm370 to be preinstalled on
every 138&148.

Note since DOS/VS and VS/1 were single virtual address space (something
like original VS2, SVS) ... E-architecture dropped the single virtual
address table into microcode ... and there were new hardware
instructions to add&remove the virtual->real address page mapping. VM370
always ran in 370 mode supporting multiple address spaces.

4341 caused lots of problems for POK ... it performed better than 3031
(erzats 158) and small cluster of 4341s outperformed 3033, cost much
less than 3033, had smaller footprint and used much less environmentals.

In 1979, I got con'ed into doing 4341 benchmarks for LLNL that was
looking at getting 70 4341s for compute farm ... sort of the leading
edge of the coming cluster supercomputing (and cloud megadatacenter)
tsunami.

It was so threatening to highend mainframes, at one point, head of POK
got allocation of critical 4341 manufacturing component cut in half.

The price, environmentals & footprint for 4300s & FBA disks had dropped
so far, that corporations started ordering large hundreds at a time for
placing out in departmental areas (inside IBM it resulted in conference
rooms becoming scarce commodity) ... sort of the leading edge of the
coming distributed computing tsunami.

Boeblingen lab had done 370 115&125 ... which was a nine position memory
bus for up to nine microprocessers ... for the 115, all microprocessors
(controllers, 370 "cpu", etc) were the same but with different microcode
loads. The 125 was identical to 115, but the microprocessor for the 370
"cpu" was 50% faster (than the other microprocessors). This
design/implementation was so threatening to other 370 models, the got
corporate to discipline Boeblingen.

At the same time that Endicott con'ed me into working on ECPS microcode
assist (for 138/148), I got con'ed into doing 125 design/implementation
which would have up to five of the faster CPU processors all in the same
machine (with four positions left for controllers). In same ways it was
as threatening to Endicott 148 as 4341 clusters was threatening to 3033.
In the escalation meetings by Endicott to kill five processor 125, I was
expected to do the technical arguments for both sides (pro/con 148+ECPS
and pro/con for 5-way 125)

-- 
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