<table cellspacing="0" cellpadding="0" border="0" ><tr><td valign="top" style="font: inherit;">hi,<br><br> We are about to start running GROMACS 4.0.4 with OpenMPI, in 8<br>nodes, quad core Rocks cluster. We made some tests, without PME and<br>found two notable things:<br><br>* We are getting the best speedup (6) with 2 nodes ( == 8 cores ). I read<br>the "Speeding Up Parallel GROMACS in High Latency networks" paper, and<br>thought that the culprit was the switch, but ifconfig shows no retransmits<br>(neither does ethtool -s or netstat -s). Does version 4 includes the<br>alltoall patch? Is the paper irrelevant with GROMACS 4?<br><br>* When running with the whole cluster ( 8 nodes, 32 cores ), top reports<br>in any node a 50% system CPU usage. Is that normal? Can it be accounted to<br>the use of the network? The sys usage gets a bit up when we configured the<br>Intel NICs with Interrupt Coalescense Off, so I'm tempted to think it
is<br>just OpenMPI hammering the tcp stack, polling for packages.<br><br>Thanks in advance,<br><br>Dra.M.Florencia Martini<br>
Laboratorio de Fisicoquímica de Membranas Lipídicas y Liposomas<br>
Cátedra de Química General e Inorgánica<br>
Facultad de Farmacia y Bioquímica<br>
Universidad de Buenos Aires<br>
Junín 956 2º (1113)<br>
TE: 54 011 4964-8249 int 24</td></tr></table><br>
<hr size=1><br><font face="Verdana" size="-2">Encontra las mejores recetas con Yahoo! Cocina.
<br>
http://ar.mujer.yahoo.com/cocina/</font>