AMBER Archive (2003)

Subject: Re: AMBER7 on linux cluster

From: Carlos Simmerling (carlos_at_csb.sunysb.edu)
Date: Fri Jun 20 2003 - 09:55:10 CDT


if you want to run it interactively, remove
the PBS stuff from your script. then
run it without PBS and see what it does.
then check the output files, including anything
output to the stdout.

how lkong does it take before you get the wallclock
exceeded error with PBS? Is it right away, or after
that amount of time (the limit)?

is it possible that you don't have enough RAM on your
cluster machines for this system size?

I think that amber7 should read amber5 files
but I am not positive- amber7 uses a new prmtop
format but for most uses it will read the old ones
as well.

Carlos

----- Original Message -----
From: "Venkata S Koppuravuri" <kvsuneel_at_linus.bmb.wright.edu>
To: "Carlos Simmerling" <carlos.simmerling_at_stonybrook.edu>
Cc: <amber_at_heimdal.compchem.ucsf.edu>
Sent: Friday, June 20, 2003 10:53 AM
Subject: Re: AMBER7 on linux cluster

> Hi Dr.Simmerling,
>
> The cluster works fine when i tried smaller structures. When I tried
> to run the job interactively the error message was
>
> --------------------------
> mpiexec: Warning: main: task 2 died with signal 15
> mpiexec: Warning: main: task 3 died with signal 15
> =>> PBS: job killed: walltime # exceeded limit #
> --------------------------
>
> I did the same run on an sgi with 2 processors it took me about 3 days.
> Then to check it on the cluster I reran it on the cluster(4nodes and
> 2ppn/node) giving the same walltime (3days) but still i get the same
> error(as above) that walltime not sufficient.
>
> 2. When the job crashes because it exceeded the runtime limit I wonder
> what happens to job directory and all the output files. It should have
> atleast some of the calculations done. When i run the job in the current
> directory rather in the TEMPDIR. I still have all the output files empty
> nothing is written to them.
>
> 3. I have done the minimization of the structure on SGI using Amber5 then
> i transfered the required files to the cluster(cluster has Amber7) to
> equilibrate it. Is it possible that the cluster is NOT doing any
> calculations because the files were generated on a different Amber
> version.
>
> Can you please help me solve the problem
>
> Regards
> -Venkat
>
>
> On Tue, 17 Jun 2003, Carlos Simmerling wrote:
>
> > can you run a non-MPI job on the cluster?
> > if so, does it work ok or also dies?
> >
> > are you using a batch scheduler? if so, sometimes
> > the error messages may go into an output file
> > created by the scheduler. run it interactively and
> > you get errors messages and also you can be sure
> > your working directory is correct, with schedulers
> > you need to set it.
> >
> >
> > ----- Original Message -----
> > From: "Venkata S Koppuravuri" <kvsuneel_at_linus.bmb.wright.edu>
> > To: <amber_at_heimdal.compchem.ucsf.edu>
> > Sent: Tuesday, June 17, 2003 12:49 PM
> > Subject: AMBER7 on linux cluster
> >
> >
> > > Hi,
> > >
> > > I am a new user of AMBER7 on a LINUX cluster. When I tried to give a
> > > simple equilibrating job to the cluster using 4 nodes and 2 pps on
each
> > > node I got the following error message
> > >
> > > ------------------------------------
> > > mpiexec: Warning: main: task 4 died with signal 15.
> > > mpiexec: Warning: main: task 5 died with signal 15.
> > > ------------------------------------
> > >
> > > All the output files are empty.
> > >
> > > Can anyone help me what the problem could be.
> > >
> > > (I did the minimization on an SGI machine using Amber5. i then
transfered
> > > the output files to the cluster and tried to do equilibration. Cluster
> > > uses AMBER7 . Hope this does not create a problem).
> > >
> > >
> > > Regards
> > > -Venkat
> > >
> > >
> >
> >
>
>