From: "Chuck Stevens" <@unisys.com>
Newsgroups: alt.sys.pdp10,alt.folklore.computers,comp.arch,comp.sys.unisys
Subject: Re: A Dark Day...
Date: Mon, 9 Jun 2003 11:47:27 -0700
Organization: Unisys - Roseville, MN

"Randall Bart" <Barticus@att.spam.net> wrote in message
news:o794evo9k8ap3i1r9lpj80gshlecaevn75@4ax.com...

> I think that Cobol-68 and Cobol-74 ran on slightly different machines.

There were actually *two* COBOL74 machines.

> In some ways the B1000 was doomed by its compatibility.  About 1981, it
> became apparent that corporate wanted to axe one of the architectures.
> There was no way that Small System or Medium Systems could replace Large
> System, so Large was safe.

Actually, I'm not convinced that was it.  The performance of the high end of
the B1900 line was quite good, even in environments using complex data
bases, in comparison with the lower end of the Large System line.  I have
heard the rumor that a fair number of upgrades from B1000 to B5900 had to be
upgraded to B6900's at considerable discount, and that at least one didn't
get satisfactory performance until Burroughs brought in a B7900.  The B1900
was surprisingly alacritous at Fortran, by the way.

The big limitation of the B1000 architecture was its two-megabyte main
memory architectural limit imposed by bit addressibility and a 24-bit
address bus.  A 32-bit (as I recall it) follow-on machine was being designed
when I visited Santa Barbara as an intern from the field in the spring of
1984, but the company'd already been burned once by the idea that "Nobody
could ever need more than 2MB on a B1000",  and bumping the address bus by a
mere 8 bits would be no more than a stopgap.  The decision was made
somewhere around that time that the follow-on machine for existing customers
of the Burroughs Small System would be the Burroughs Large System, and
efforts to make sure that the Large System would do all that the B1000 would
do (things like CANDE page mode, the SORT compiler, the PASS command, SMCS
capabilities in COMS, etc.,) date from this period as I remember.

I don't think the B1000 was ever really seriously considered as a Medium
System replacement because, as I understand it, at a hardware level, both
the Small and Large systems were "process driven" machines, handling
interrupts when they bloody well got around to it, whereas the Medium System
was very much an "interrupt driven" machine.  While a fully-pumped 2-meg
1990 with 5n disk and a spare disk channel or three *might* be able to give,
say, a B3900 a run for its money on batch and on-line processing, as I
remember putting check sorters on the B1900 brought just about everything
else to a crawl, and trying to do much else at the same time a sorter run
was in process usually resulted in lots of too-late-to-pocket-select
rejects.   As long as you didn't try to do much multiprogramming during a
sorter run, everything was fine.   Medium System folks expected to be able
to pound their terminals and print their reports and (later) have their
customers safely withdraw money from their ATM's while the checks were being
sorted online.  I don't think the B1000 was quite up to that for any but the
smaller banks.

Interestingly, though, they did market a box called the Reader Sorter
Processor for use on the Large Systems (which also didn't do on-line
reader-sorter work well, for the reason that it responded when it got around
to responding).  That box was a (very) thinly-disguised B1720, which
basically did nothing other than run the reader-sorter program.  I never
actually saw one in production, and don't know how it interacted with the
Large Systems.  The reason the B1000 could work in that environment was
precisely that it was a dedicated machine; there wasn't anything competing
for its attention besides the Reader Sorter itself.

    -Chuck Stevens