[gmx-users] Scaling Problems

[Dallas Warren]

Lynne,

>Did you try recompiling lam-mpi with the latency optimisations that were 
>suggested (?by David) some time ago on the list?

I certainly did.  I went back, recompiled to lam-6.5.6 using 
-with-tcp-short=524288 versus 64k default, --with-rpi=usysv

For those interested, here are the details for the system along with the 
scaling numbers and details:

System:
         Intel P733 Dual Processor
         100 base Ethernet
         Linux Redhat 7.3
         lam-6.5.6 (tcp-short=524288)
         fftw-2.1.3 (enable-mpi)
         gromacs-3.1.4 (enable-mpi)

MD Simulation:
         7.5 nm cube
         16 molecules, 16 Na, + water
         42,000 atoms
         500 steps
         0.002 ps step
         PME (order=4, 0.12nm fourier spacing)

grompp switches:
         -sort
         -shuffle
         -np #

Initiation of mdrun:
         mpirun C mdrun -s .....
         1 CPU run time approximately 500 sec

SCALING
Cluster tcp=64k tcp=524k        PME off         PME(6/0.17nm)

1 box/2 CPU     71%             72%             86%             84%
2 box/2 CPU     54%             56%             83%             74%
2 box/3 CPU     40%             41%             73%             62%
2 box/4 CPU     27%             30%             71%             51%

So:
         # changing the tcp-short of lam did improve things slightly, but 
not really that much to justify the right royal pain it was ;-)
         # PME doesn't scale very well, as is noted in the manual etc.
         # PME does scale better when the fourier grid spacing is increased 
and the PME order increased

I am going to see if there is much of a difference for a longer run, to 
reduce the effect of the setup and shut down period that doesn't 
parallelise, as noted by Anton.

Catch ya,

Dr. Dallas Warren
Research Fellow
Department of Pharmaceutical Biology and Pharmacology
Victorian College of Pharmacy
Monash University
dallas.warren at vcp.monash.edu.au
+61 3 9903 9076
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20020919/7b044a76/attachment.html>