[mpi3-coll] October Telecon

[Torsten Hoefler]

Hello Collectives-WG,
I propose

   October 9th, 12:00pm EDT

as the time/date for our monthly collectives workgroup teleconference.

Please let us know if there are any (strong) objections to this time. I
will reserve the spot this Thursday if there are no objections.

We will mainly be talking about the feedback we got from the whole forum
on our proposals. Christian's and my notes are at:
https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/forum090408

We will also try to bring everybody up to speed who could not attend the
last forum!

Im particular we have to decide/discuss (those points also serve as
agenda items):

1) talk about function interfaces (how do they look like)
  - we decided for the "calls for everything" way
  - now we should flesh out details how the newly proposed collectives
    should look like
  - I will keep track of our decisions and prepare a full proposal
    document
2) Discuss progress on the subitems
  a) non-blocking collectives interface will be defined (see
  straw-votes)
   * make final decisions about the following items
   - mix blocking/non-blocking calls
   - use use MPI_Requests
   - multiple outstanding colls
   - nest/mix collectives
   - prefix (I/A/N/...)
   - max outstanding colls?
   - how do we want to support hardware acceleration?
   - any other details to discuss here?
  b) discuss research progress on sparse collectives
   - we have the Octopus EuroPVM paper
     http://www.unixer.de/publications/#hoefler-europvm08-qm
   - further directions?
   - volunteers?
  c) discuss research progress/evaluation of persistent colls
   - directions to go?
   - volunteers?
3) Updates on topological collectives
  a) figure out which operatiosn make sense and how to call them
   - neighbor exchange (alltoall?)
   - neigbor reduce 
   - neighbor allgather
   - neighbor allreduce
   - versions for cart comms with distance > 1?
   - more?
  b) discuss applications using them (neighbor exchange is clear, but
     the others?)
   - need to get in touch with real users to decide anything useful
4) Updates on MPI Plans 
   - keep it going (research-wise)
   - finish prototype implementation
5) Variable size collectives (does anyone pick this topic up?)
   - Alexander/Hans-Joachim?
   - do we have an application/implementation to look at?
   - any news?
6) MPI-2.2 issues - revisited 
   a) fix non-scalable graph interface 
    - will do implementation
    - discuss issues (if any)
   b) local reduction operation (needed by libraries, e.g., LibNBC)
    - any issues?
   c) local progress function (caused heavy discussions in Forum)
    - issues ;)?
   d) request completion callbacks (for better progression, e.g., LibNBC)
    - discuss what to allow/disallow during callbacks
    - do we need to limit anything?
   e) partial pack/unpack (needed by libraries, e.g., LibNBC)
    - do we want to continue this one?
7) other items (please post here!)

Thanks & Best,                                                                  
 Torsten Hoefler                                           

-- 
 bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ -----
Torsten Hoefler       | Research Assistant
Open Systems Lab      | Indiana University    
150 S. Woodlawn Ave.  | Bloomington, IN, 474045, USA
Lindley Hall Room 135 | +01 (812) 855-3608