[mpi3-coll] July/August telecon
moody20 at llnl.gov
Mon Jul 28 14:57:09 CDT 2008
The two key areas I'd like to work with our users on here at LLNL are
non-blocking and neighbor-exchange (topological) collectives. I could
perhaps also look into the benefits of persistent collectives. Our
users would like to start using non-blocking collectives asap -- they
are excited about this. I suspect we'll be able to show some real value
for the neighbor-exchange collectives on our clusters, as well. And
given IB's need to pin memory, we may also be show performance gains
with persistent collectives.
For non-blocking collectives, should I start with libNBC? Where can I
get the latest copy of it? I know you have done some work with
neighbor-exchange collectives, too. Do you have a similar library for that?
Torsten Hoefler wrote:
> July 31, 12:00pm EDT
>as the time/date for our monthly collectives workgroup teleconference.
>Please let us know if there are any strong objections.
>We will mainly be talking about the feedback we got from the whole forum
>on our proposals. My notes are at:
>Im particular we have to decide (those points also serve as agenda
>1) "One call fits all" vs. "Calls for everything:
> - we should make a decision on which model we want to use in order to
> be able to flesh out semantic details of the operations
>2) Do we want/need all combinations that are semantically possible?
> - the Forum accepted the usefulness of non-blocking collectives as proved
> - the Forum wants to see more research/use cases for sparse collectives
> - the Forum wants to see more research/use cases for persistent colls
> - who wants to invest time into this?
>3) Updates on topological collectives
>4) Updates on MPI Plans (I have a slightly different
> proposal/implementation than Christian for the same thing)
>5) Variable size collectives (does anyone pick this topic up?)
>6) MPI-2.2 issues (I will add the WG's proposals to the MPI-2.2 wiki so
> that this can serve as a base for discussion)
> a) fix non-scalable graph interface (obvious)
> b) local reduction operation (needed by libraries, e.g., LibNBC)
> c) local progress function (caused heavy discussions in Forum)
> d) request completion callbacks (for better progression, e.g., LibNBC)
> e) partial pack/unpack (needed by libraries, e.g., LibNBC)
>7) other items
>Please post any additional proposals items to discuss to the list!
>Thanks & Best,
> Andrew Lumsdaine and Torsten Hoefler
More information about the mpiwg-coll