MPI (lack of) Performance?

Martin Konold konold@alpha.tat.physik.uni-tuebingen.de
Mon May 10 22:41:32 1999


On Mon, 10 May 1999, Jonathan King wrote:

> performance.  My results (using the simple ping-pong MPI program at the
> bottom) inside of MPI are only order 23Mb/s.

Doing tcp/ip is expensive and not well suited to high performance MPI. We
measured for example about 24 MB/s with MPI over tcp/ip on Myrinet but 
95MB/s with the Score 0 copy implementation of ScoreMPI using the very
same hw. Similar numbers can be achieved with SCI.

Simply count the number of data copy operations and context switches.

Regards,
-- martin

// Martin Konold, Herrenbergerstr. 14, 72070 Tuebingen, Germany  //
// Email: konold@kde.org                                         //
Anybody who's comfortable using KDE should use it. Anyone who wants to
tell other people what they should be using can go to work for Microsoft.
KDE not only OpenSource[TM] but OpenMinded[TM].
KDE: A stable GUI for a reliable OS.


 | To unsubscribe, send mail to Majordomo@cesdis.gsfc.nasa.gov, and within the
 |  body of the mail, include only the text:
 |   unsubscribe this-list-name youraddress@wherever.org
 | You will be unsubscribed as speedily as possible.