[RegCNET] Parallel computing problem

[Guangyu Wu]

Dear All,

I am running RegCM3.1 parallel version on a PC SUSE
Linux cluster. I installed Mpich1.2.7 with Intel
FC9.0, SSH and chp4 configuration. Then I compiled
regcm using the default Makefile (Makefile_CL2, remove
gmlib). My network is a 100M ethernet. I share regcm
and ICBC files on the mater node to slave nodes
through NFS.

While using a 48x48 grid, I found the computing time
for 4 nodes is almost same as 2 nodes. When I changed
the domain size to 72x72 with same other
configurations, the 4 nodes cluster is faster than two
nodes, but almost same with 6 nodes.

I am wondering it is a problem related to network and
mpich setting, or this is a problem associated with
regcm code itself. 

Hope someone can help me figure out a solution. Thanks
very much in advance.




 
____________________________________________________________________________________
Sucker-punch spam with award-winning protection. 
Try the free Yahoo! Mail Beta.
http://advision.webevents.yahoo.com/mailbeta/features_spam.html