1) Persistent collective operations are needed to enhance optimizability.
These have been removed from MPI-2.
1a) A Transpose-type algorithm, both intra- and inter-communicator, is
needed. [various discussions back to September, most offline]
1b) Other collective ops drpped from MPI-2 might need to go here.
2) Generalized requests with persistent form, seem needed here too.
We need to resolve what pieces of MPI-2 left on the floor by the forum
need to be picked up.
3) Get semantics for one-sided, receiver-based communication in MPI.
4) Making progress a QOS property for all non-blocking operations must
be discussed.
5) We need an atomic handoff of buffers in event handlers for event-driven model.
6) We need a way to clarify time-based with multiple buffers period, per
Arkady.
7) We need to consider removing priority from channel-based and introducing
into regular sends (late binding) as another type of tagging.
[Return to an earlier proposal by Jerrell/Steve]
8) We still need to consider sets of channels and the implied resource
object, for discussing transitions of resources. This might be more general
than the new proposal to unify init/modify/delete, brougth up this week
at the meeting.
I am sorry that I was sick and missed Forum RT discussion this week;
See you in San Diego.
-Tony