very slow of the parallel jobs

Questions regarding the compilation of VASP on various platforms: hardware, compilers and libraries, etc.


Moderators: Global Moderator, Moderator

Post Reply
Message
Author
hu007

very slow of the parallel jobs

#1 Post by hu007 » Mon May 22, 2006 7:22 pm

Dear VASP users,

We are running parallel jobs on a cluster, but the jobs doesn't scale up with the number of processors used. It
runs very slow. Does anybody konw how to fix this problem?
Is the problem that Lam MPI doesn't work near as well on these systems as the MPI versions designed for these computers (derivatives of MPICH MPI)? Thanks!
Last edited by hu007 on Mon May 22, 2006 7:22 pm, edited 1 time in total.

tjf
Full Member
Full Member
Posts: 107
Joined: Wed Aug 10, 2005 1:30 pm
Location: Leiden, Netherlands

very slow of the parallel jobs

#2 Post by tjf » Tue May 23, 2006 10:50 am

LAM is designed for clusters and works very well. At least as well as MPICH.

What's your interconnect? How are you setting up the parallelisation?

(Consider also OpenMPI, where the LAM development has now moved along with several other MPI implementation teams.)
Last edited by tjf on Tue May 23, 2006 10:50 am, edited 1 time in total.

admin
Administrator
Administrator
Posts: 2921
Joined: Tue Aug 03, 2004 8:18 am
License Nr.: 458

very slow of the parallel jobs

#3 Post by admin » Tue May 23, 2006 1:12 pm

I fully agree with forum member tjf.
Apart from that, please mind that there is a small part of VASP (about 2-3% of the code) which cannot be parallelized, therefore VASP will not scale exactly linearly with the number of processors. The performance in general is best for about 16 parallel processors.
Last edited by admin on Tue May 23, 2006 1:12 pm, edited 1 time in total.

Post Reply