Friday, 12 August 2011

Performance results for PMI-optimised version


Slightly better than it used to be.

Parallel Efficiency:
cpu_cores: 1    E = 99.5%
cpu_cores: 2    E = 96.6%
cpu_cores: 4    E = 91.7%
cpu_cores: 8    E = 79.7%
cpu_cores: 16   E = 75.3%
cpu_cores: 24   E = 67.6%

And simulation total against node activation:


Posted by at

3 comments:

  1. Unknown13 August 2011 at 15:16

    Hi!
    May I ask what you mean by "New PMI"?

    ReplyDelete
  2. Anonymous14 August 2011 at 18:00

    Looking better. Got any multiple node timings at all?

    adrianj

    ReplyDelete
  3. Megatelevizor14 August 2011 at 21:36

    Hi Olaf,
    Glad you're still following this blog. By "new PMI" I meant your PMI module with optimised invoke method: I made it use MPI.Gather instead of MPI.gather to avoid doing any pickling. Here's a discussion thread on the MPI4PY google group where the difference between the two routines is explained: http://groups.google.com/group/mpi4py/browse_thread/thread/a1741896517c7d12/5c5e604047365348?lnk=gst&q=konstantin#5c5e604047365348.

    The difference in usage between the optimised invoke and the standard one is that the send and receive buffers need to be passed together with the function call. I wanted to use the Bcast routine as well (instead of bcast), but it turned out to be a bit more complicated and in any case broadcasting data doesn't take much time in my simulations.

    ReplyDelete

Subscribe to: Post Comments (Atom)