[mpiwg-p2p] Next meeting 18th Jan 2016

Jeff Hammond jeff.science at gmail.com
Mon Dec 14 10:19:35 CST 2015


Can someone tell me the best replacement for the following pattern if
freeing active requests is made erroneous?

void send_fire_and_forget_ack(int target, MPI_Comm comm)
{
  MPI_Request request;
  MPI_Isend(NULL, 0, MPI_BYTE, target, special_tag, comm, &request);
  MPI_Request_free(&request);
  return;
}

The reason to not use MPI_Send here is it can block on resource exhaustion.

Note that this is a real usage model.  One of my applications send
fire-and-forget acks and it was a performance issue on Blue Gene/Q.
Because we send a payload - it is unclear if this is actually necessary - I
replaced MPI_Send with MPI_Bsend to solve the problem (yes, Bsend really
does help here), but we could instead use Isend+Request_free (which would
be easy since our payload is trivial and could be a global const).

The reason I bring this up even though Bsend works for us is that I would
much rather deprecate Bsend that make freeing active requests erroneous.

Jeff

On Mon, Dec 14, 2015 at 7:55 AM, Daniel Holmes <dholmes at epcc.ed.ac.uk>
wrote:

> Hi all,
>
> The next point-to-point teleconference will be Monday 18th January 2016 at
> 11am Central US via webex.
>
> I will send out another reminder nearer the time.
>
> Tentative agenda:
>
> MPI-4.0, issue #11: add communicator assertions (remember implications for
> tools!)
> MPI-4.0, issue #26: deprecate cancel for send requests
>
> MPI-4.0, issue TBD: add ARECV/FSEND
> MPI-4.0, issue TBD: deal with minor correction to MPI_WAIT raised by Bill
> and Rajeev (a la Bill's email 29/06/2015)
> * MPI-4.0, issue TBD: deprecate request free for active receive requests
> (arising from Dec 2015 meeting)*
> MPI-4.0, issue TBD: define batch requests (a la Pavan's email of
> 14/12/2015)
>
> * MPI-Next, issue TBD: disallow request free for active receive requests
> (i.e. upgrade advice to erroneous)*
> MPI-Next, issue #27: disallow cancel for send requests (i.e. upgrade
> advice to erroneous)
>
> Cheers,
> Dan.
>
> --
> Dan Holmes
> Applications Consultant in HPC Research
> EPCC, The University of Edinburgh
> James Clerk Maxwell Building
> The Kings Buildings
> Peter Guthrie Tait Road
> Edinburgh
> EH9 3FD
> T: +44(0)131 651 3465
> E: dholmes at epcc.ed.ac.uk
>
> *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
> mpiwg-p2p at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-p2p
>



-- 
Jeff Hammond
jeff.science at gmail.com
http://jeffhammond.github.io/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-p2p/attachments/20151214/bcd78940/attachment-0001.html>


More information about the mpiwg-p2p mailing list