Re: [Condor-users] Condor_compile fails to link MPI object files

[Tanzima Zerin Islam <tz.islam@xxxxxxxxx>]

Thanks Greg, that was helpful. I guess, what made me try condor_compile in the first place was that MPICH-V ppl have provide a package of condor's checkpoint library to enable their mpich compiler produce checkpointable MPI executable. However, they might not be using condor's checkpoint library out of the box ... I don't know for sure...

Thanks anyways,

--Tanzima

On Thu, Apr 8, 2010 at 10:34 AM, Greg Thain <gthain@xxxxxxxxxxx> wrote:

Tanzima Zerin Islam wrote:

Hi all,
 To follow up on what I have done to link 3 compiled files into 1 checkpointable MPI executable:
It is the "IS" Nas Par benchmark application written in c.

1. cc -g  -o setparams setparams.c
2. cc -g  -c -I/tmp/NPB3.3/NPB3.3-MPI/common    is.c
3. cc -g  -c -I/tmp/NPB3.3/NPB3.3-MPI/common    c_print_results.c
4. cc -g  -c -I/tmp/NPB3.3/NPB3.3-MPI/common    c_timers.c

I'm sorry, but the condor checkpointing technology is unable to checkpoint processes that have inter-process communication of most any kind.  It can't checkpoint MPI codes today.

If all the MPI nodes are going to run on the same machine, you might want to investigate the DMTCP checkpointing libraries on source forge.

Sorry, and good luck,

-Greg