Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!att!tut.cis.ohio-state.edu!ucbvax!unisoft!fai!sequent!dafuller
From: dafuller@sequent.UUCP (David Fuller)
Newsgroups: comp.benchmarks
Subject: bc a la MIMD
Keywords: bc concurrent parallel benchmark sequent symmetry s81
Message-ID: <49425@sequent.UUCP>
Date: 28 Dec 90 18:52:50 GMT
Organization: Sequent Computer Systems, Beaverton, OR
Lines: 41

OK, just to belabor the point and because I'm having a snit over all
of these fast unis reporting results, I decided to make my own rules
and see how many bcs I could run concurrently on one of our lab machines:

Machine - Sequent Symmetry S81, 16 processors, 20MHZ 386/387 system:

Instances 	Real Time
-------------------------
    1		23.2
    8   	23.4
    16		23.4
    32		44.8
    64		88.5
    128		176.6
    256		352.7

I ran these in a simple-minded fashion by simply spawning children with
a shell script and waiting for all processes to complete.  

It is an imperfect test because I should really do a little more work and
set all the processes at a barrier and turn them loose at once; 
the 256 process test took about 10 seconds to start up so the rampup and
taildown aren't as abrupt as they should be.  

Still, the test shows razor-flat linearity once you go past the number of
CPUs on the system, indicating that the OS didn't do much to slow things
down.  

If you want to argue throughput, then I completed one bc every 23.4/16 
seconds, or about 1.5 seconds per bc :-).  I suspect that if I put the other
4 CPUs I have lying around into the machine I could do 23.4/20 per bc with
an ultimate limit of 23.4/30 seconds per bc if I had the max of 30 CPUs.

Dave

My comments are my own.  My tongue firmly in cheek.
-- 
Dave Fuller				   
Sequent Computer Systems		  Think of this as the hyper-signature.
(708) 318-0050 (humans)			  It means all things to all people.
dafuller@sequent.com