<html>
<head>
<style>
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Verdana
}
</style>
</head>
<body class='hmmessage'>
Dear axel<br><br>Thank you very much for your help!<br><br>Just following your suggestion, I export OMP_NUM_THREADS=1, the calculation finish within several seconds!<br><br>Thank you very much for your help! <br><br>The machine I used is SGI-Altex-4700. After I download the code, I just do <br>1) configure<br>2) make all<br><br>I donot know if I still other set or configurations to optimize my calculation in SGI-Altex-4700?<br><br>Thanks<br><br>Xiangang Wan<br>Department of Physics<br>Nanjing University<br>P.R. China<br><br>> Date: Mon, 10 Nov 2008 21:55:58 -0500<br>> From: akohlmey@cmm.chem.upenn.edu<br>> To: lo_wan_2005@hotmail.com<br>> Subject: Re: [Pw_forum] Why my test-calculation is so slowly? SGI4700 for Si take about 1 hours<br>> <br>> On Tue, 11 Nov 2008, Lo_wan_2005XW wrote:<br>> <br>> XW> <br>> XW> Dear PWscf users<br>> <br>> dear xw,<br>> <br>> please don't forget to specify your full name and affiliation.<br>> <br>> you are a victim of not overly smart sysadmins and the strange<br>> choices of intel. you are most likely linking to MKL and as of<br>> version 10.0 it automatically multi-threads across _all_ available<br>> cpu 'cores'. on an SGI altix this has disasterous consequences<br>> for the performance. you have to set OMP_NUM_THREADS=1 and<br>> it will be _much_ less of a problem (check out the mailing list<br>> archive for multiple occurances of this "feature".<br>> <br>> please not that the first test examples don't overly scale<br>> as well (for testing, you may be better off with the contents<br>> of the tests directory using only 2-4 processors.<br>> <br>> cheers,<br>> axel.<br>> <br>> p.s.: if your sysadmins are not totally ignorant, please<br>> ask them to at export OMP_NUM_THREADS=1 to the default<br>> environment, and they'll avoid a lot of problems with<br>> jobs not running overly efficient.<br>> <br>> XW> <br>> XW> I just download the code and install it in our SGI-4700 without any warnning or error.<br>> XW> <br>> XW> After that, I try to do the example-calculation (example01).<br>> XW> <br>> XW> My machine is SGI-Altex-4700, so I change the the following line in the file of run_example <br>> XW> <br>> XW> <br>> XW> <br>> XW> $PW_COMMAND < si.scf.$diago.in > si.scf.$diago.out<br>> XW> <br>> XW> <br>> XW> <br>> XW> to <br>> XW> <br>> XW> <br>> XW> <br>> XW> $PARA_PREFIX $PW_COMMAND < si.scf.$diago.in > si.scf.$diago.out<br>> XW> <br>> XW> <br>> XW> <br>> XW> I set <br>> XW> <br>> XW> <br>> XW> <br>> XW> PARA_PREFIX="mpirun -np 12"<br>> XW> <br>> XW> <br>> XW> <br>> XW> After that, I type ./run_example<br>> XW> <br>> XW> I can see the following in the screen:<br>> XW> <br>> XW> <br>> XW> <br>> XW> sgi4700:~/Quantum-Espresso/espresso-4.0.3/examples/example01> ./run_example <br>> XW> <br>> XW> /disk2/Quantum-Espresso/espresso-4.0.3/examples/example01 : starting<br>> XW> <br>> XW> This example shows how to use pw.x to calculate the total energy and<br>> XW> the band structure of four simple systems: Si, Al, Cu, Ni.<br>> XW> <br>> XW> executables directory: /disk2/Quantum-Espresso/espresso-4.0.3/bin<br>> XW> pseudo directory: /disk2/Quantum-Espresso/espresso-4.0.3/pseudo<br>> XW> temporary directory: /disk2/tmp<br>> XW> checking that needed directories and files exist... done<br>> XW> <br>> XW> running pw.x as: mpirun -np 12 /disk2/Quantum-Espresso/espresso-4.0.3/bin/pw.x -npool 4<br>> XW> running bands.x as: mpirun -np 12 /disk2/Quantum-Espresso/espresso-4.0.3/bin/bands.x -npool 4<br>> XW> <br>> XW> cleaning /disk2/xgwan/tmp... done<br>> XW> running the scf calculation for Si...<br>> XW> <br>> XW> <br>> XW> <br>> XW> <br>> XW> <br>> XW> <br>> XW> <br>> XW> By using "top", I can find there are 12 pw.x. My machine take about 2 hours to finish this simple calculation. It is quite strange! <br>> XW> I compare the my output and the data saved in referece-directory, it shows that my result, like total energy, is right comparing with the referenced data.<br>> XW> <br>> XW> Attached pleased find my output data.<br>> XW> <br>> XW> I will appreciate if someone can kindly tell me the possible reason for it. <br>> XW> <br>> XW> Thanks<br>> XW> <br>> XW> XW<br>> XW> <br>> XW> _________________________________________________________________<br>> XW> MSN ÖÐÎÄÍø£¬×îÐÂʱÉÐÉú»î×ÊѶ£¬°×Áì¾Û¼¯ÃÅ»§¡£<br>> XW> http://cn.msn.com<br>> <br>> -- <br>> =======================================================================<br>> Axel Kohlmeyer akohlmey@cmm.chem.upenn.edu http://www.cmm.upenn.edu<br>> Center for Molecular Modeling -- University of Pennsylvania<br>> Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323<br>> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425<br>> =======================================================================<br>> If you make something idiot-proof, the universe creates a better idiot.<br><br /><hr />使用新一代 Windows Live Messenger 轻松交流和共享！ <a href='http://im.live.cn/' target='_new'>立刻下载！</a></body>
</html>