I cannot attend. My trac acct is fixed though. Will upload material for discussion by others. Jeff Sent from my iPhone

On Apr 13, 2014, at 5:24 AM, Daniel Holmes <[email protected]> wrote:

Hi all,

The next point-to-point teleconference will be on Monday 14th April 2014 at 11am Central US via webex.

Connection details are on the p2p wiki.

Agenda: 1) Streams: I have some notes - see below - from a discussion last week with Bill Gropp, Torsten (et al from the EPiGRAM project) 2) Receive_reduce 3) Arecv 4) AOB (= any other business)

Cheers, Dan.

Streams ======= QCDOC developers want persistent channel semantics; actually use one-sided, multi-value variables Breaking message boundaries enables pipe-lining - makes it fundamentally different to RMA? Libraries exist to do this - what are they? Streaming is well defined in OpenSPL Papers from the MPI-2 era (about RMA) suggest writing papers on this and getting feedback Reach out to other communities, e.g. SKA - Square Kilometre Array Should this be added to MPI or added as a library on top - does this need small enabling changes Bill suggests we craft a broad-visibility paper Affects the discussion on Arecv, which would be handled by streams? Getting MPI to manage/allocate memory (as in Arecv) is dangerous and bad

Flow-control ============ May just be remote notification and a sequence of messages that remove the need for local notification - reduces protocol messages (from 3 to 1 per message) Integral part of streaming - turns global flow-control problem into local buffer management May address load-imbalance? Hmm

-- Dan Holmes Applications Consultant in HPC Research EPCC, The University of Edinburgh James Clerk Maxwell Building The Kings Buildings Mayfield Road Edinburgh, UK EH9 3JZ T: +44(0)131 651 3465 E: [email protected]

*Please consider the environment before printing this email.*

The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

_______________________________________________ mpiwg-p2p mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-p2p