Dear MPI Forum Colleagues:

The Persistent WG in collaboration with the Point-to-point WG plans to read a 
new proposal entitled "Partitioned Communication."


A brief explanation follows since this is new to many of you as a formal 
proposal for MPI-4:

MPI currently provides all multi-threaded functionality through the existing 
interface that was designed primarily for single-threaded operation. There are 
no interfaces designed specifically to address the unique requirements of 
multi-threaded communication in MPI.

Partitioned communication modes extend the behavior of persistent 
point-to-point communication using a non-local initialization call. Partitioned 
communication is "Partitioned" because it allows for multiple contributions of 
data to be made from potentially multiple threads/tasks to a single 
communication operation.

This is the proposal formerly known as finepoints, which we've been working on 
for several years.

Ryan Grant will lead this reading in Chicago.

Tony Skjellum

Anthony Skjellum, PhD

Professor of Computer Science and Chair of Excellence

Director, SimCenter

University of Tennessee at Chattanooga (UTC)  [or]

cell: 205-807-4968

mpi-forum mailing list

Reply via email to