2 Mar
2017
2 Mar
'17
12:17 a.m.
On Wed, Mar 1, 2017 at 2:45 AM, Ronald Natalie <ron(a)ronnatalie.com> wrote:
AIX/370 was a real product.
Indeed it was.
All of these AIX versions came from the same source
code and used the
IBM TCF to allow you to transparently run executables across nodes in the
cluster.
Exactly right. TCF - Transparent Computing Facility -- No mean trick...
you can mix PS/2 and 370 in the cluster, so root on desk allowed me root
on the mainframe too. What was cool was that the TCF will look at the
executable and find the proper CPU. The big mistake was that that node id
was stored in a single 32 bit word and assigned per bit, which was a
scaling issues.
I was at Locus Computing Corp (aka LCC or just "Locus"), who developed AIX
for IBM under contract and TCF was part of it. The direct result of the The
LOCUS Distributed System Architecture
<https://www.amazon.com/Distributed-System-Architecture-Computer-Systems/dp/0262517191/ref=sr_1_1?s=books&ie=UTF8&qid=1488391384&sr=1-1&refinements=p_27%3AGerald+J.+Popek>
from
UCLA. The book actually describes much of the AIX/370 work, but starts
with the original UCLA work. I did not work on the IBM project, although a
number of my peers did. I was higher to help developed TNC - Transparent
Network Computing, which is was used in Intel's Paragon and DEC's
TruClusters and a never shipped HP Cluster Product. Many of the same ideas
but we wanted a separate team that never saw the IBM code so there could
never be any concern about ownership. The architects like me and Roman,
were allowed to talk to the AIX architects, such as Bruce; but we keep
separate development environments at separate sites. After the IBM work
ended, all of the Locus distributed system folks the struct around went to
work on TNC and the technology go sold off and licensed. What was
interesting is that TNC was open'ed sourced after the Compaq/HP mergers and
put into Linux but I've forgotten the URL (I'll search and follow up).
It's a real shame it never went anywhere. It was a very, very cool.
The only AIX that didn’t play was the completely
independent (and in
my opinion somewhat brain damaged) IBM/RT UNIX. If there was a TCF-based
RT kernel, I never saw it, even inside the IBM labs.
That was IBM politics. LCC has the contract for the original AIX port to
the 370. When the RT was developed, the Austin team was ramped up. One
of our members of the TUHS list who is remaining silent I see is not saying
why but I know was there ;-) and might known the actual politics, I never
did. But when the AIX/RT port was forked, they started with AIX/370 code
base and removed the TCF code. But LCC still had the AIX/370 contract
from Enterprise system group to maintain AIX/370. And also, Locus had the
contract from Entry Systems, who all they wanted TCF. So AIX/386 and
AIX/370 as Ron points out were one code base, one dev team (at LCC in
California).
Dan Cross said: "I had understood was that AIX/370 was actually OSF/1
based"
It maybe that by the end, the user space was based on the OSF/1 user space
code. That was true for HP and DEC also. But I can definitely state
AIX/370 and AIX/386 were one set of source trees and all of was done by
Locus Computing Corporation certainly through the mid 1990s.
Clem