[Pw_forum] Too many data communications between nodes when run pw.x in parallel?

Wang Qinjing 041019019 at fudan.edu.cn
Mon Dec 29 18:05:01 CET 2008


Dear all:

I noted that if I made calculation by pw.x in paralllel, there were too many data communications between nodes. That was attributed to the data communication of wave functions. There is a severe delay to get information from these nodes.

If I set disk_io='low', things are much better. But the main node (or the first node) still has problem in net communication. I noticed a charge density file is still written after each scf step. Is that the problem?  Or is there anything I can do to improve this situation?

Thanks in advance.

Yours Wang

------------------------------------
王秦镜,复旦大学物理系研究生
Wang Qinjing, Physics Department, 
Fudan University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20081230/03ff49e8/attachment.html>


More information about the users mailing list