benchmark info.

Carl S. Gutekunst csg at pyramid.pyramid.com
Thu Jan 19 18:14:59 AEST 1989


>I am looking for some benchmark info. ($$, MIPS, architecture etc) on
>the sequent and pyramid machines.  Any pointers in that direction will
>be much appreciated.

Both companies have benchmark groups, and will be more than happy to discuss
numbers with you. Both will also be able to tell you why the other's numbers
are a pack of lies. :-) Unfortunately, I have often found that the companies
that profesionally write and run benchmarks also have strong personal biases;
so you can't rely on those either. 

If you really want benchmarks, a good reference is John Mashey's UNIX Perfor-
mance Characterization Suite. Far better is to pick something you use a lot,
and design your own benchmark around it. Assuming you are at all serious about
purchasing a machine, again both companies will be happy give you time on
their machines to try them out.

Keep in mind that the Pyramid 9000 and the Sequent Symmetry, though both sym-
metric multi-processors, are conceptually very different machines. At the very
least, Pyramids use fewer and larger CPUs; so non-parallizable processes will
run faster, but you have a higher incremental upgrade cost. Pyramids are also
more targeted towards commercial environments, while Sequents are targeted
towards academic, engineering, and scientific.

And of course, these aren't the only two players in the supermini market: CCI,
Arix, AT&T, DEC, Encore, Alliant, Convex. And you might be surprised at the
capabilities of high-end "micros" from Sun, NCR, or MIPS. Also remember that
the market is not static; don't look just at what the company has today, but
also what they are planning on for six months, a year, two years, five years
down the road. Demand answers, and be satisfied if you can at least get some
educated waffling. :-) On the other hand, be very careful when asking about
what the company can ship today, not what they are beta testing next week.

When I buy machines, I try establish a matrix of priorities versus machines,
and mark down which machines will meet each priority. This is a big help in
eliminating emotional biases. Of course, whether or not the applications you
need run on that particularly machine should be very high on the matrix. 

<csg>



More information about the Comp.sys.pyramid mailing list