AMBER Archive (2007)

Subject: AMBER: sander.MPI vs MPICH2 problem!!!

From: Swarup Gupta (swarupgupta_at_yahoo.com)
Date: Fri Mar 23 2007 - 20:33:44 CST

I have installed the sander.MPI successfully. I am using MPICH2. It has been installed on a HP-cluster and Redhat linux is the operating system. The processors are Opteron. Each node has two processors.
   
  When I submit the sander.MPI job (say Job-1) from the master node to run it on one or more nodes to run the jobs on multiple processors I do not see to run the sander.MPI programs on any of the CPUs. Suppose I submit the job-1 from masternode to run on ‘node-1’, just after submission I see that python2.2 is running node-1 and also sander.MPI starts to run and then with a fraction of a seconds sander.MPI disappears (also python) and nothing more happens. But at the same time the job is not aborted from the master node. I do not get any output file created by sander. So I guess that sander is not actually executed and not even aborted. Does it imply that I have some problem with MPI? (the non-mpi version of sander runs perfectly on the same machine). The jobscript is OK since it works on another cluster.
   
  If I just run (without running with mpi) the command $AMBERHOME/exe/sander.MPI then it creates “mdout” and gives the following message:
   
  “[unset]: aborting job:
  application called MPI_Abort(MPI_COMM_WORLD, 1) - process 0”
   
  In this case it creates an output file and I guess that amber installation is OK but there is some communication problem from MPICH2. If anybody has any idea please let me know.
   
   
  Thanks
  SG

 
---------------------------------
Don't pick lemons.
See all the new 2007 cars at Yahoo! Autos.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu