Page 1 of 1

Scalapack CPU use on cluster - linking ATLAS

PostPosted: Sun Jan 16, 2011 2:45 pm
by jsrassa
Hello, I would thank anyone who can help me with this problem. I made a Bewoulf cluster using Ubuntu with two computers and I'm trying to use Scalapack but the CPU use is really low (25% most of the time) so its taking more time than just using Lapack. I'm trying to link with the ATLAS library but it still does not solve the problem. The way I am compiling is:
mpif90 -lscalapack-openmpi -lblacsF77init-openmpi -lblacsCinit-openmpi -lblas-openmpi -llapack_atlas -lf77-blas -lcblas -latlas -lpthread
Thank you very much

Re: Scalapack CPU use on cluster - linking ATLAS

PostPosted: Fri Jan 21, 2011 12:01 pm
by admin
HI,
Which routines are you using, what are the matrix sizes?
Could you tell more about your machine? Freq?

Re: Scalapack CPU use on cluster - linking ATLAS

PostPosted: Sun Jan 23, 2011 7:03 pm
by jsrassa
Hello. I'm using these routines: pdgemm pdgemv, pdgetri and pdgetrf. The principal matrix sizes I have tried are 200x300 and 80x150 (different processes). I'm currrently using two computers and a 100 Mbit router for the cluster. Both computers have two processors each. One of them has an AMD vision processor and the other one has an AMD turion.
Thank you

Re: Scalapack CPU use on cluster - linking ATLAS

PostPosted: Mon Jan 24, 2011 12:38 pm
by admin
You should try with larger matrices. (limit should be around 5000x5000 depending on your machine)
Best would with to increase your matrix size from 500x500 up to the limit and plot the time at each run. You would see the scaling effect.

Re: Scalapack CPU use on cluster - linking ATLAS

PostPosted: Fri Jan 28, 2011 10:21 am
by jsrassa
Thank you for your help, it looks like the problem was that I was using very small size.