AMBER Archive (2000)

Subject: Re: MPI Sander on a 5 node SMP cluster

From: David Konerding (dek_at_cgl.ucsf.edu)
Date: Tue Dec 05 2000 - 14:52:36 CST


"Dr Chris L. Brown" writes:
>--============_-1236129112==_ma============
>Content-Type: text/plain; charset="us-ascii" ; format="flowed"
>
>Hi
>
>We have just assembled a 5 node 10 CPU ( PIII 800 ) SMP cluster using
>redhat 6.2.
>
>We are using MPICH 1.2.1 (P4 server with comm=shared) which works
>fine with all nodes running the test programs without a problem (eg
>pi3 etc with the /share/machine.linux files setup).
>
>However, when we try and submit a job to sander we get a error
>
>> mpirun -np 10 /usr/local/amber6/exe/sander etc. etc.
>
> unknown flag:
>-p4pg

MPICH uses a few extra flags to pass information to the application.
p4pg tells where the p4 procgroup file (automatically generated by mpirun)
is. Normally AMBER ignores the p4pg flag (see mdfil.f).

I would suggest you check your mdfil.f to make sure it skips p4pg.
Amber-6.0 should skip it, do you have a "test" version of AMBER6 or did
you change the source code in any way?

Dave