Jeff paraphrased an unnamed source as suggesting that: "any MPI program that relies on a barrier for correctness is an incorrect MPI application." . That is probably too strong.

How about this assertion?

If there are no wildcard receives - every MPI_Barrier call is semantically irrelevant.

It is the exception that tests the rule.

If someone can provide an example of an MPI_Barrier that is required by an application based on MPI communication and that does not use wildcard receive I am interested in seeing it. I do not know of a counter example but also do not have proof of the assertion I place before the group.

No fair using examples with non-MPI interactions among tasks or with job steering by asynchronous triggers from outside the job. I can construct them myself.

MPI_WIN_FENCE is semantically required in some situations and examples that show a semantic need for MPI_WIN_FENCE do not count against the assertion.

I have appreciated the descriptions from Gus, Asjley and others of some non-symantic justifications for an MPI_Barrier.

Dick Treumann - MPI Team
IBM Systems & Technology Group
Dept X2ZA / MS P963 -- 2455 South Road -- Poughkeepsie, NY 12601
Tele (845) 433-7846 Fax (845) 433-8363