On 8/26/13 10:31 AM, "Dries Kimpe" <[email protected]> wrote:

* Barrett, Brian W <[email protected]> [2013-08-26 16:06:00]:

[...]

...

That being said, I don't think resource exhaustion corner cases are a deal breaker for me. I think some implementation-dependent phrasing might be acceptable. It might be useful to define the message transmission/matching semantics for these corner cases. For example, if MPI_Arecv returns an error because of resource exhaustion, is the message lost (my preference) or left in the receive queue? If the message is lost, what happens to the sender if the send was a synchronous send?

I think allowing the message to be 'lost' should be last resort and not taken lightly, since as far as I know, a message cannot be lost using the existing MPI functions (ignoring fault tolerance/resilience here) and, as you pointout in your example, might cause many complexities at other places.

Sure they can. The standard is totally silent on what happens when MPI_RECV returns an error. You're firmly in the realm of undefined behavior. Brian -- Brian W. Barrett Scalable System Software Group Sandia National Laboratories