Re: [OMPI devel] Latency optimizations

[George Bosilca]

I have another send side optimization on the pipeline. As Galen will take care of the receiver side, I can focused on the send side. As far as I see, this will rip off few tens of microsecond, bringing our latency on shared memory to basically the same level as MPICH2. This might requires some small changes in the way the BTLs handle the send operation, but the changes will be really minor. I'll send an RFC around soon.

  george.

On Feb 28, 2008, at 3:36 PM, Richard Graham wrote:

FYI, About six months ago several of us spent some time coming up with a plan to deal with the latency problems in Open MPI. George went ahead and has been implementing the send side changes of this optimization over the last several months, but has not had time to get to the receive side. Galen is picking up on this, and will be checking in changes over the next several weeks. The gist of these is going from an active-message tag approach with one tag per protocol (ptp, one-sided, etc) to an 8 bit global tag space, and finer grain functions (short message, rendezvous packet, ...), as well as

some function consolidation.

Rich

_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

smime.p7s
Description: S/MIME cryptographic signature