We’ve been doing work specifically on this. In the latest alpha, an experimental version of the ULFM specification was added. You’re welcome to try it out. If you’re not familiar, I’d recommend reading through some of the documentation on www.fault-tolerance.org. You can find some tutorials along with the specification.

Thanks,
Wesley


On Dec 10, 2014, at 6:31 AM, YANG Fan <iddmbr@gmail.com> wrote:

Hi,

Is it possible for an MPI distributed cluster to continue working if one node dies? I'm not sure if MPICH provides such functionality.

It seems that MPI_Comm_create requires that all processes in the superset communicators to be alive; while the errhandler with --disable-auto-cleanup also does not avoid such issue, as one process cannot call MPI_Finalize().

Thanks in advance!

Best Regards,
Fan
_______________________________________________
discuss mailing list     discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss