Re: Is non-blocking collective I/O desirable?
Steve Huss-Lederman (firstname.lastname@example.org)
Mon, 17 Feb 1997 04:09:35 -0600 (CST)
I am sorry to jump in late but I am running around Europe. One point
I would like to raise is that just because generalized requests have
(or will be I hope) worked out does not imply to me that nonblocking
collective are also solved. GR have restrictions on when they make
progress that is likely to be different with collective operations.
Another point is that it was raised you could do nonblocking with
threads. Part of what makes GR hard is that we are not willing to
assume that threads exist in an MPI implementation. Thus, saying you
can do it with threads does not completely solve the problem.