>I suppose the next thing is to try this with some other MPI
implementation,
>just to try to narrow down the possible errors.
...
>you could do this if you have access to some other cluster.
Here's what I have so far.
I get the same thing on a 64-bit cluster running similar MPI
and interconnect except that it goes for 13000 steps rather
than 12000.
But, if I run on two processors on the same node, the program
goes to completion.
So, it's likley related to the node interconnect.
I've written Myricom to see if they have any ideas. We get
the error messages I printed earlier pretty frequently. Maybe
there's some configuration we could change.
Let me know if I can do anything else.
:-) Lachele
B. Lachele Foley, PhD, '02
Assistant Research Scientist
Complex Carbohydrate Research Center, UGA
706-542-0263
lfoley.uga.edu
Received on Wed Apr 05 2006 - 23:49:41 PDT
