Thanks for sharing, very interesting history to me.  You guys were pros.. particularly amazing to me how far ahead the machine abstractions were on the various IBM machines (CP, S/38, VRM) compared to most of the industry.

On Wed, Feb 19, 2020 at 11:44 PM Charles H. Sauer <sauer@technologists.com> wrote:

> On Feb 18, 2020, at 7:41 AM, Kevin Bowling <kevin.bowling@kev009.com> wrote:
>
> ...
>
> IBM abandoned the idea of any ukernel with AIX3 for RISC/6000.. Charlie may be able to add commentary on that but it was almost certainly for performance which was paramount in the workstation wars and RS6K had an front runner opening.
>

I initially missed Kevin's ping after my spam filter put several TUHS messages in /var/mail/devnull. (I eventually skim subject lines of messages that go there.)

I could write more than I want to/should about how the VRM came to be and not to be, but will try to add a little to what I've said before (https://notes.technologists.com/notes/2017/03/08/lets-start-at-the-very-beginning-801-romp-rtpc-aix-versions/). I'm trusting 30+ year-old memories here and not looking at the various papers and manuals that might inform.

I joined Glenn's AFWS project July 5, 1982. There was no well defined software plan yet. Glenn wanted to do something useful and significant, and proposed that we do the VRM. We had several distinct user environments in mind. I took the lead in writing a specification of the VMI (virtual machine interface) while others started prototyping. We were way overly ambitious with abstractions along the lines of the single level store of (Glenn's) System 38, trying to take advantage of the 40 bit addressing of the Rosetta virtual memory chip, yet still heavily influenced by CP/CMS. After a few months, Al Chang, primary person behind CP.R, came to Austin for a design review of what we'd done. He told Glenn he'd grade our work "C+". That might have been generous.

We scaled back our ambitions dramatically, started working with ISC. About the time (1983) of the transition from "ad tech" to "product" organization, it became clear that our virtual memory manager needed to be scrapped and we lifted what Al had done for CP.R and put it in the VRM.

In hindsight, the VRM turned out better than it might have. Besides AIX there was a version of Pick for VRM that sold about 4000 copies according to https://en.wikipedia.org/wiki/IBM_RT_PC. Though the VMI cost us some in performance, we were surprisingly successful in minimizing the penalties. But with AIX 3 and RS/6000 we wanted to take dramatic steps forward, and it made no sense to preserve the VMI.

Anecdotal comments on other TUHS/COFF discussions:

If I recall correctly, pcc, eventually including the HCR optimizing phase, was bundled with base AIX. Initially, the C compiler based on the PL.8 compiler would only run on CMS, so it was not generally available outside of IBM, but app vendors, especially CAD vendors, were enabled and encouraged to come to Austin to use it to get the best performance. The native C compiler based on PL.8 compiler concepts ended up being a complete rewrite, outside of Yorktown, and sold as a separate product.

Producing software products, getting them released, priced, etc. was very confusing to me most of the time I was at IBM. Part of it was the history that Clem has cited. Part of it was confusion about the antitrust suits against IBM. Part of it was confusion about whether and what software was patentable. Academics and others wanted access to the modeling & simulation software, RESQ, my team developed at Yorktown. Eventually, the concept of "Research Distributed Program" was agreed upon and RESQ was the first instance: https://technologists.com/sauer/RA144.pdf. However, we were forced to price RESQ much higher than I thought reasonable. I had already transferred to Austin by the time the release was official -- I don't know how many copies were sold. But source code was necessary to take full advantage of RESQ so the PL/I source was included on the tapes.

When OSF was announced, with the intention of making AIX source available to the other OSF companies, I was stunned because it was so uncharacteristic of the IBM I thought I knew. It would be interesting to know how that would have worked out if OSF had stuck with AIX and IBM had delivered the source on the schedule everyone hoped for, but that's on a different timeline than this one.


--
voice: +1.512.784.7526       e-mail: sauer@technologists.com           
fax: +1.512.346.5240         web: https://technologists.com/sauer/
Facebook/Google/Skype/Twitter
: CharlesHSauer