efficiency of parallelization?


[ Molcas user's WWWBoard ]

Posted by Darko Babic on October 13, 2002 at 19:16:29:

Hi!

I have installed Molcas parallelized with MPICH-1.2.4,
all compiled with Portland 4.0. However, I don't get
much (if anything) from the parallelized code. I
didn't make extensive testing: so far I have only
results from a few RASSCF optimization and frequency
calculations (extending trough 7-8 hours) which are
certainly not representative.

Anyway, before I spend more CPU-time with testing,
could someone share an experience in the gain achieved
by parallelization?

I see a speedup with SEWARD, but MCKINLEY spends the
same time (in my case about 3 hours) on 1, 2 (SMP) and
6 (3 x 2 SMP) processors. All processors show equal
load during the calculation.

We have a Rocks/RedHat 7.2 cluster of dual Athlons
1800+, with RAID0 IDE disks (giving up to 80 Mb/sec
in read) and 100 Mbit Ethernet on the inner side.

Thank you!

Darko Babic


Follow Ups:



Post a Followup

Name:
E-Mail:

Subject:

Comments:


[ Follow Ups ] [ Post Followup ] [ Molcas user's WWWBoard ]