[GE users] Tight Integration of MPICH with SGE

Waseem Ahmad Waseem.Ahmad.1 at Sun.COM
Tue Jul 27 22:39:15 BST 2004


Here is how things are being done. A bash shell script, which is 
submitted to SGE(EE), calls the perl script.The perl script in turn 
calls mpirun multiple times with different number of processes and 
combines the results.
Presently, I am able to run any MPI job over SGE with MPICH being 
loosely configured. But, with tight integration i am only able to run 
the sample script provided in the mpi directory of SGE.
The latest problem that i am having with tight integration is that SGE 
is reporting some errors

error: executing task of job 141 failed:
Broken Pipe

  while at the same time the MPICH is also reporting errors

p4_error: Child process exited while making connection to remote process 
on host1: 0

Please let me know what is happening here. Note that if i specify only 
two processes the benchmark runs fine even with tight integration. 
Problems occure only for proc specification of more than two.
Just as side note the cluster consists of 8 machines. 6 of them are dual 
processor machines while remaining two are single processor machines.

thanks.
waseem



Reuti wrote:
> What exactly is your question - how to access the environment variables in your 
> Perl script and put them in the mpirun call?
> 
> $my_nslots = $ENV{NSLOTS};
> $my_tmpdir = $ENV{TMPDIR};
> 
> 
>>I think the MPI Jobs spawned through the perl script are unable to get 
>>$NSLOTS. cat $TMPDIR/machines gives me the list of nodes in the cluster 
>>and thus looks to be fine.
>> What should i do to pass these variables to MPI jobs.
>>thanks.
> 
> 
> Cheers - Reuti
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
> 
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list