Hi Thomas,
> the TI process communication is independent of ambers normal
> parallelisation scheme, but it certainly sounds as if it couldnt hurt to
> try a different MPI implementation (are you using the LAM that came with
> amber 10?).
This is an information only our sysadmin can give me and he is on vacation. :-)
>
> Also, check every one of your output files, sometimes error messages
> appear only in one of the two TI processes or in the queueing output.
> Assuming your cluster has 4 CPU per node, if the 4 CPU run goes through
> fine, maybe the problem only occurs when you run on more than one node and
> be network-related?
>
We have 16 CPU´s per node. But we are using infiniband anyway, so we are getting
a better performance with distributing on many nodes than just using one nodes
whole CPU´s. It doesn´t seem to be a network problem.
Best regards,
Hannes
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 10 2009 - 01:09:33 PDT
