AMBER Archive (2005)

Subject: FW: AMBER: mpirun sander8 problem

From: S. Frank Yan (SYan_at_gnf.org)
Date: Thu Feb 24 2005 - 02:41:14 CST


Hi all,

I ran many test cases. When single CPU is used (mpirun -np 1), sander
runs fine. But when two CPUs are used, sander hangs right before:

| Atom division among processors:
| 0 1271

Only the sentence:

Number of triangulated 3-point waters found: 0

is in the output file.

The processes go on though without any additional output and it doesn't
stop. Any idea what might have caused this problem?

Thanks,
Frank

> -----Original Message-----
> From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On
Behalf
> Of Carlos Simmerling
> Sent: Wednesday, February 23, 2005 5:05 AM
> To: amber_at_scripps.edu
> Subject: Re: AMBER: mpirun sander8 problem
>
> does the mpi code work for a single processor? are you sure that the
> input you are using works properly? mpi runs do not always give the
> proper error message if the error occurs on a cpu other than the
master.
>
> S. Frank Yan wrote:
>
> >Hi,
> >
> >I was able to recompile MPI using gcc and ifort with ch_shmem and
> >compiled sander8 on a 2-CPU Linux box. I tested it with the dhfr
> >testing case. The program runs in parallel but it stops spitting out
> >output after "3. ATOMIC COORDINATES AND VELOCITIES... " section. I
can
> >see the CPU usage is around 100% for both CPUs, but sander seems to
be
> >stuck in a loop. Using mpirun -v yields "running ../../exe/sander on
2
> >LINUX smp processors". Does anyone have the same experience like
this?
> >
> >Thanks a lot.
> >Frank
> >
> >
> >
> >
>
-----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu