now, I set $HOME as shared directory, but when doing ompi-checkpoint, it shows:(nimbus1 is the remote machine in
my cluster)

[nimbus1:12630] opal_os_dirpath_create: Error: Unable to create the sub-directory (/home/mpiu/ompi_global_snapshot_1662.ckpt/0) of (/home/mpiu/ompi_global_snapshot_1662.ckpt/0/opal_snapshot_4.ckpt), mkdir failed [1]
[nimbus1:12630] Error: No metadata filename specified!

why is that?

cheers
fengguang

On Tue, Mar 23, 2010 at 10:37 AM, Fernando Lemos <fernandotcl@gmail.com> wrote:
On Tue, Mar 23, 2010 at 12:24 PM, fengguang tian <fernyabc@gmail.com> wrote:
> Hi
>
> I am using open-mpi and blcr in a cluster of 3 machines, and the checkpoint
> and restart work fine in single machine,but when doing checkpoint in
> clusters environment, the ompi-checkpoint hangs

Besdies what has been said in another thread (regarding 1.4 and
checkpointing to shared directories), you might want to make sure your
app is terminated if you send a SIGTERM to it. Some apps might ignore
SIGTERM or handle it in a way that does not cause the apps to quit.

ompi-checkpoint --term is simply ompi-checkpoint + sending SIGTERM to
the application (not sure whether SIGTERM is sent to each process
individually or not).
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users