5 Mar
2013
5 Mar
'13
10:43 a.m.
On 03/05/2013 10:34 AM US Central Time, Jim Dinan wrote:
You should ignore my comments on MPICH FT. My info is clearly out-of-date. It sounds like what you're looking for should be fully supported. :)
Well, almost :-). Some things could not be done cleanly while staying within MPI-3. For example, when you do a wildcard receive, it will always return an error if any process in the communicator is dead. The MPI Forum is working on fixing this by allowing the user to "opt in" for this wildcard stuff. I'm a few months behind on the MPI-3.1 fault-tolerance proposal, but I believe this is still present. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji