<div dir="ltr"><br><br><div class="gmail_quote">On Fri, Sep 19, 2008 at 1:31 PM, <span dir="ltr"><<a href="mailto:hessb@mpip-mainz.mpg.de">hessb@mpip-mainz.mpg.de</a>></span> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<br>
<br>
> What -ddorder cartesian does?<br>
<br>
This determines the order of the PP and PME nodes.<br>
If the XT4 would really support cartesian communication this would<br>
be a lot faster, but unfortunately it does not.<br>
Still it will probably be faster when all the PME nodes are physically<br>
close toghether. -ddorder cartesian will do this.<br>
-ddorder pp_pme will also work well, but I expect cartesian<br>
to be slightly faster, you can try both.<br>
</blockquote><div><br>For one of my test systems interleave is fastest on XT4. But the difference is very small (5%). My system with 120k atoms scales to 160 cores (with 48 cores for PME) giving 21.1ns/day. XT4 latency is really high. So that in some cases it is 50% slower than on low latency clusters (Myri 10G).<br>
<br>Roland <br></div></div><br><br clear="all"><br>-- <br>ORNL/UT Center for Molecular Biophysics <a href="http://cmb.ornl.gov">cmb.ornl.gov</a><br>
</div>