[Wien] k-point parallel job in distributed file system

[Joachim Luitz]

XU ZUO schrieb:
> Unfortunately, I am suffered from the instability of the k-point
> parallelization. I understand that this problem is caused by the bad NFS
> performance (problems on read/write latency and synchronization) and that
> adjusting $delay and $sleepy may solve the problem. However, as the cluster
> load and traffic are dynamic, it is better to design adaptive code, which
> can handle this problem dynamically. 
>   
then you could try to use the parameter

extrafine:1

in the .machines file, which then will distribute chunks of single 
k-points. Thus it will dynamically adapt to your cluster's load, but 
will create extra network traffic when summing up.

Regards
  Joachim

--
luitz.at | interfacing art, science and technology

Dipl.-Ing. Joachim Luitz KEG 
Wohlmuthgasse 18 . A-3003 Gablitz . T +43 2231 612540 . Fax +43 2231 612544
buero at luitz.at . http://www.luitz.at . skype://jluitz