I still think that the pml fast path fixes would be good.

Sent from my PDA.  No type good.

----- Original Message -----
From: devel-bounces@open-mpi.org <devel-bounces@open-mpi.org>
Sent: Tue Mar 17 18:23:18 2009
Subject: [OMPI devel] OMPI vs Scali performance comparisons

A colleague of mine ran some microkernels on an 8-way Barcelona box (Sun
x2200M2 at 2.3 GHz).  Here are some performance comparisons with Scali.
  The performance tests are modified versions of the HPCC pingpong
tests.  The OMPI version is the trunk with my "single-queue" fixes...
otherwise, OMPI latency at higher np would be noticeably worse.

              latency(ns)   bandwidth(MB/s)
            (8-byte msgs)   (2M-byte msgs)
            =============    =============
      np    Scali    OMPI    Scali    OMPI

       2      327     661     1458    1295
       4      369     670     1517    1287
       8      414     758     1535    1294

OMPI latency is nearly 2x slower than Scali's.  Presumably, "fastpath"
PML latency optimizations would help us a lot here.  Thankfully, our
latency is flat with np with the recent "single-queue" fixes...
otherwise our high-np latency story would be so much worse.  We're
behind on bandwidth as well, though not as pitifully so.
devel mailing list