Request Handlers, please

Carter Edwards : carter@ticam.utexas.edu (carter@ticam.utexas.edu)
Fri, 10 May 1996 12:44:08 -0500 (CDT)

Please post the minutes of the April 24-26, 1996 meeting.
The discussion and vote to remove Request Handlers from
the MPI2 spec. is of particular interest.

It was noted in Chapter 6, External Interfaces, that
Request Handlers will reappear in a TBD Chapter on Threads.

--

I propose that Request Handler semantics be "orthogonal" to Thread semantics. An MPI user should be able to use Request Handlers without having to be concerned with Thread semantics.

A simple suggestion to achieve this "orthogonality": If Threads are present, then the execution of a Request Handler preempts the Thread from which it was posted. A Request Handler could either execute within its posting Thread or block its posting Thread until the Request Handler "returns".

If a MPI User wants a particular Handler or set of Handlers to execute in a separate Thread then the MPI User must first create a Thread and then post the Handler(s) within that Thread.

--

This baits a larger discussion regarding MPI-2 and Threads. If an MPI implementation internally creates Threads then should these threads: 1) be transparent to the MPI user and 2) not introduce consistency problems in the presence of an MPI user's Threads.

---------------------------------------------------------------------------- Carter Edwards, carter@ticam.utexas.edu, http://www.ticam.utexas.edu/~carter ----------------------------------------------------------------------------