[molpro-user] parallel IO issue? (fwd)
Rika Kobayashi
rxk900 at anusf.anu.edu.au
Wed Oct 12 04:40:33 BST 2011
Our System Manager has picked up the following. I haven't had a chance to look at it
but thought I'd throw it into the ether for comments.
Seth, we might need more details about the job from you (e.g. jobtpe and molpro version).
Machine details can be found at http://nf.nci.org.au/facilities/vayu/hardware.php.
>
> Hi Seth,
>
> This job
>
> vayu3:~ > nqstat 320238
> %CPU WallTime Time Lim RSS vmem memlim cpus
> 320238 R sco564 n62 mgc43s3. 30 35:11:02 48:00:00 163GB 175GB 180GB 16
>
> has slowed to a crawl and the reason appears to be that all processes are
> writing to (overwriting) the same file:
>
> /short/m03/Arylmethanes/Malachite-Green/Isomer/c43s3.5.pvdz/mg.c43s3.5.pvdz.c87_orbital_87.1.cube
>
> Even if all the processes are writing the same data, its not clear you'll
> get a meaningful file out of this. And having multiple nodes write to the
> same locations in a file is a disaster for performance. Is this "standard"
> Molpro behaviour?
>
> David
>
More information about the Molpro-user
mailing list