Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] RFC: sm Latency
From: Richard Graham (rlgraham_at_[hidden])
Date: 2009-01-22 14:13:59


Nothing wrong with worrying about ping-pong, this is the first thing people
look at.

The question about the unexpected queue was to check and see if proper MPI
matching is done. If there is matching entry in the unexpected queue, this
is the one that is the correct match from the MPI matching rules.

Rich

On 1/22/09 12:51 PM, "Eugene Loh" <Eugene.Loh_at_[hidden]> wrote:

> Richard Graham wrote:
>> Re: [OMPI devel] RFC: sm Latency In the recvi function, do you first try to
>> match off the unexpected list before you try and match data in the fifo¹s?
>>
> Within the proposed approach, a variety of things are possible.
>
> Within the specific code I've put back so far, I happen to check the
> unexpected list. If it has something on it, the "immediate" receive reverts
> to the traditional code path. If the unexpected list is empty, I proceed on
> to the FIFO. Again, that's just how the code works that I've put back so far
> to the workspace mentioned in the RFC.
>
> You can let me know why you ask and what behavior you recommend.
>
> Again, to be quite honest, a lot of my interest here is motivated by
> (ping-pong style) benchmarks. I feel dirty just admitting that, but then I
> remind myself of all the reports I see of people getting really bad latencies
> using OMPI.
>
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel