[mvapich-discuss] Buffer size of unexpected receive queue, is there a way to increase the buffer?

[balaji at mcs.anl.gov]

Just as a FYI -- the below code creates zero unexpected messages. Everything is expected. And it's a perfectly synchronous application, so there's not much pressure on the internal buffers either (though the performance might not be as great).

 -- Pavan

----- "Craig Tierney" <Craig.Tierney at noaa.gov> wrote:

> I am trying to figure out a way to increase the
> unexpected receive queue, if it is even possible
> or necessary.
> 
> I have code that looks a bit like:
> 
> for i=0,npes-1
>    msgsize=Some function to determine how much data (even 0) should be
> sent to another PE
>    mpi_send(send to process i, msgsize)
>    if (msgsize > 0)
>      mpi_send(send to process i, data of msgsize bytes)
> done
> 
> for i=0,npes-1
>    mpi_recv(receive from procss i, msgsize)
>    if (msgsize > 0)
>      mpi_recv(receive from process i, data of msgsize bytes)
> done
> 
> On my job with 864 cores, the code is blocking in here.
> I believe that since the receives are not posted, that I
> am most likely hitting some internal buffer limits.  This
> code works on other systems, but they have had to tweak
> internal buffer sizes to get it to work.
> 
> I cannot change the code in any way.  So I am looking
> for how the size of the unexpected receive queue is set
> and if/how it can be changed.
> 
> Thanks,
> Craig
> 
> 
> -- 
> Craig Tierney (craig.tierney at noaa.gov)
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss