[TUHS] UNIX on Tandem

[Brad Spencer via TUHS]

Tom Lyon via TUHS <tuhs at tuhs.org> writes:

> Ran across this citation today.  Anyone ever seen this paper?  Known of the
> system?   I'm pretty sure they're talking about Tandem(TM), not just
> generic tandem.
>
> A Distributed UNIX System-The Tandem Experiment. A. M. Usas, Proc Natl
> Electron Conf, 34 (October 1980), pp 16-8.


I don't know if my experiences with the Tandem are what is being talked
about in that paper, but given its apparent date, I suspect I was after
it, but here is what I remember about using a Tandem.

Some version of the Tandem was white boxed at AT&T as the Starserver (or
Star Server or some such) in the late 1990s time frame.  I believe that
it came to be used in the product I was on for possibly two reasons, the
first was it might have been mandated from higher up and the second is
that the RBOC customers wanted "fault tolerance" or at least some
computer system that behaved a bit like some of the switching gear.  The
Tandem that was used was certainly FT in most ways.  The backplane was
an active beast where boards could be pulled out and replaced while the
system was running and it would continue to run.  Nearly every board
type, from CPU, to memory to disk controllers to power supplies had at
at least one spare and the load would be moved to a spare if the system
detected a fault.  All that you saw was a console message indicating
that a board was removed or failed or whatever.  The Unix that ran on it
was a modified SVR3, I believe, and the processor was some sort of MIPS,
if my memory is still correct.  The group I was in had mostly the white
boxed Starserver version which was a 120v / 240v sort of thing.  But
Tandem did make a central office version that ran on DC and we did have
one of those because one of the RBOC customers wanted to put our system
in a real central office which ran on DC mostly.  The systems continued
to be reliable until they got very old and the complicated backplane
started to get quirks in it.  The backplane was one component that
really didn't have a fault tolerate part to it, or at least not enough
of one and when the backplane got bad enough the system was pretty much
done and over with.

A lot of the software groups at 6200 Broad St. used the Starserver /
Tandem for the products that they worked on in the same time frame.
Later, the RBOC customers decided that FT wasn't worth the cost of the
systems and we moved the product to HP with a brief diversion over to
NCR (or GIS or whatever NCR was called after it was bought) for a bit.






-- 
Brad Spencer - brad at anduin.eldar.org