On 2/2/07, <b class="gmail_sendername">Shi Jin</b> <<a href="mailto:jinzishuai@yahoo.com">jinzishuai@yahoo.com</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
> There is a point which is not clear for me.<br>><br>> When you run in your shared-memory machine...<br>><br>> - Are you running your as a 'sequential' program<br>> with a global,shared<br>> memory space?
<br>><br>> - Or are you running it through MPI, as a<br>> distributed memory<br>> application using MPI message passing (where shared<br>> mem is the<br>> underlying communication 'channel') ?<br>
<br>Thank you for replying.<br>I run the code on a shared memory machine through MPI,<br>just like what I do on a cluster. I simply did:<br>petscmpirun -np 18 ./code<br><br>I am not 100% sure whether MPICH-2 will automatically
<br>use shared memory as the underlying commnunication<br>channel instead of the network but I know most MPI<br>implementations are smart enough to do so (like<br>LAM-MPI I used before). Could anyone confirm this?<br>Thank you.
</blockquote><div><br>This is missing the point I think. It is just as Satish pointed out.<br>Sparse matrix multiply is completely dominated by memory bandwidth<br>and the shared memory machine has contention between the processes.
<br>I guarantee you that the performance problem is in the effective memory<br>bandwidth per process.<br><br> Matt<br></div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Shi<br><br><br><br><br>____________________________________________________________________________________<br>Sucker-punch spam with award-winning protection.<br>Try the free Yahoo! Mail Beta.<br><a href="http://advision.webevents.yahoo.com/mailbeta/features_spam.html">
http://advision.webevents.yahoo.com/mailbeta/features_spam.html</a><br><br></blockquote></div><br><br clear="all"><br>-- <br>One trouble is that despite this system, anyone who reads journals widely<br>and critically is forced to realize that there are scarcely any bars to eventual
<br>publication. There seems to be no study too fragmented, no hypothesis too<br>trivial, no literature citation too biased or too egotistical, no design too<br>warped, no methodology too bungled, no presentation of results too
<br>inaccurate, too obscure, and too contradictory, no analysis too self-serving,<br>no argument too circular, no conclusions too trifling or too unjustified, and<br>no grammar and syntax too offensive for a paper to end up in print. -- Drummond Rennie