jcduell_at_lbl_dot_gov
Date: Fri Apr 23 2004 - 10:30:58 PDT
On Fri, Apr 23, 2004 at 11:36:59AM -0500, Kevin wrote:
> Dear All,
>
> When we use blcr to checkpoint the MPI program, the current pid of
> mpirun is needed. But in fact, it is not possible that we ask all the
> programmers who code mpi applications print out the pid of mpirun in
> theire program. So it there some suggestion to get the pid of mpirun
> without modifying the user MPI application source codes, while we are
> ready to use blcr's command line functionality?
Tingyu,
The normal case will evenutally be for our checkpoint/retart software to
be integrated into the batch system on a machine (like PBS,
LoadLeveller, etc.), so you'd use a utility like 'qstat' to get your
job's ID, and then you could checkpoint/restart it with that ID.
We are adding such support to the Scalable Software Systems system
project's software:
http://www.scidac.org/ScalableSystems/
And we hope at some point to see OpenPBS support too (and support from
commercial vendors).
--
Jason Duell Future Technologies Group
<jcduell_at_lbl_dot_gov> Computational Research Division
Tel: +1-510-495-2354 Lawrence Berkeley National Laboratory