<html>
<head>
<style>
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Verdana
}
</style>
</head>
<body class='hmmessage'>
Hi,<br><br>Setting -npme 2 is ridicolous.<br>mdrun estimates the number of PME nodes by itself when you do not specify -npme.<br>In most cases you need 1/3 or 1/4 of the nodes doing pme.<br>The default -npme guess of mdrun is usually not bad,<br>but might need to tuned a bit.<br>At the end of the md.log file you find the relative PP/PME load<br>so you can see in which direction you might need to change -npme,<br>if necessary.<br><br>Berk<br><br>> Date: Fri, 9 Jan 2009 18:37:37 -0700<br>> From: nsapay@ucalgary.ca<br>> To: gmx-users@gromacs.org<br>> Subject: Re: [gmx-users] Best performace with 0 core for PME calcuation<br>> <br>> Nicolas a écrit :<br>> > Hello,<br>> ><br>> > I'm trying to do a benchmark with Gromacs 4 on our cluster, but I <br>> > don't completely understand the results I obtain. The system I used is <br>> > a 128 DOPC bilayer hydrated by ~18800 SPC for a total of ~70200 atoms. <br>> > The size of the system is 9.6x9.6x10.1 nm^3. I'm using the following <br>> > parameters:<br>> ><br>> > * nstlist = 10<br>> > * rlist = 1<br>> > * Coulombtype = PME<br>> > * rcoulomb = 1<br>> > * fourier spacing = 0.12<br>> > * vdwtype = Cutoff<br>> > * rvdw = 1<br>> ><br>> > The cluster itself has got 2 procs/node connected by Ethernet 100 <br>> > MB/s. I'm using mpiexec to run Gromacs. When I use -npme 2 -ddorder <br>> > interleave, I get:<br>> Little mistake: I used the wrong cluster specifications. There is 4 <br>> cores per nodes and they communicate with Infiniband.<br>> > ncore Perf (ns/day) PME (%)<br>> ><br>> > 1 0,00 0<br>> > 2 0,00 0<br>> > 3 0,00 0<br>> > 4 1,35 28<br>> > 5 1,84 31<br>> > 6 2,08 27<br>> > 8 2,09 21<br>> > 10 2,25 17<br>> > 12 2,02 15<br>> > 14 2,20 13<br>> > 16 2,04 11<br>> > 18 2,18 10<br>> > 20 2,29 9<br>> ><br>> > So, above 6-8 cores, the PP nodes are spending too much time waiting <br>> > for the PME nodes and the perf forms a plateau. When I use -npme 0, I <br>> > get:<br>> ><br>> > ncore Perf (ns/day) PME (%)<br>> > 1 0,43 33<br>> > 2 0,92 34<br>> > 3 1,34 35<br>> > 4 1,69 36<br>> > 5 2,17 33<br>> > 6 2,56 32<br>> > 8 3,24 33<br>> > 10 3,84 34<br>> > 12 4,34 35<br>> > 14 5,05 32<br>> > 16 5,47 34<br>> > 18 5,54 37<br>> > 20 6,13 36<br>> ><br>> > I obtain much better performances when there is no PME nodes, while I <br>> > was expecting the opposite. Does someone have an explanation for that? <br>> > Does that means domain decomposition is useless below a certain real <br>> > space cutoff? I'm quite confused.<br>> ><br>> > Thanks,<br>> > Nicolas<br>> ><br>> ><br>> > _______________________________________________<br>> > gmx-users mailing list gmx-users@gromacs.org<br>> > http://www.gromacs.org/mailman/listinfo/gmx-users<br>> > Please search the archive at http://www.gromacs.org/search before posting!<br>> > Please don't post (un)subscribe requests to the list. Use the <br>> > www interface or send it to gmx-users-request@gromacs.org.<br>> > Can't post? Read http://www.gromacs.org/mailing_lists/users.php<br>> <br><br /><hr />See all the ways you can stay connected <a href='http://www.microsoft.com/windows/windowslive/default.aspx' target='_new'>to friends and family</a></body>
</html>