[mpich-discuss] Latency or B/W, which is the bottleneck for most MPI codes?

[junli gu]

   Hi everybody:


   I wonder to most MPI codes, which is the performance bottleneck, the
message latency or the B/W?

  I am doing research to optimize the cache coherent shared memory
architecture (CMP) to support message passing workloads more efficiently. I
am trying to reduce the latency for message data, however later I realized
that maybe the B/W is the real bottleneck. I have heard the comments that
most well-written MPI codes  are not latency-sensitive but B/W sensitive. If
this is true,  instead of reducing the latency for single message, making
sure large chunk of data transferring faster(larger B/W) is the right way to
improve performance.

 Please feel free to give some comments about this. If someone farmilar with
the well-written legacy MPI codes can give me some suggestions, I'll find it
useful.

  Thank you in advance.

  Junli

-- 
************************************************
Junli Gu--谷俊丽
Coordinate Science Lab
University of Illinois at Urbana-Champaign
************************************************
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100305/9ae6b3d9/attachment.htm>