Multiprocessing?

Kay Diederichs Kay.Diederichs at uni-konstanz.de
Wed May 8 06:36:18 EST 1996


I have made the same experience on the university computing centre 10-processor
R8000 PowerChallenge, which mostly runs at a load average of around 30. When I
performed the tests (benchmark1 and benchmark2), the system was at IRIX 6.0.1.

If I only look at the CPU time reported at the end of the job, X-PLOR appears
to run slightly faster if more than one processor is used. However, the wall
clock time actually increases by a factor of 3 up to 10 (probably depending on
the load). I never had the opportunity to test on an otherwise idle
PowerChallenge.

What I now do is use NUMTHREADS of 1 and run several jobs at the same time (if
it makes sense for the project). Due to the large size of the program and data
area, X-PLOR gets most of the CPU cycles. Using one processor, the wall clock
time is about half of that required on our SGI R4400 200MHz workstation.

Pretty sobering multiprocessing experience.
-- 
Kay Diederichs                    
email: Kay.Diederichs at uni-konstanz.de     Tel: +49 7531 88 4049  Fax: 3183
Fakultaet fuer Biologie, Universitaet Konstanz 
Postfach 5560 (M656), D-78434 Konstanz, Germany




More information about the X-plor mailing list