[Pw_forum] Ways to reduce disk i/o when doing phonon calculations with ph.x
Hans Lind
halin at ifm.liu.se
Tue Aug 2 20:21:45 CEST 2011
Ok, thanks for the answer.
> It depends. If you have calculations involving "dozens" of atoms, you
> can perform a
> single irrep/wave-vector calculation on each node, working on local
> disks. Then, you
> have just to perform a final run that "collects" all the pieces. The
> latter involves only
> partial dynamical matrices, i.e. a small amount of data. The
> infrastructure to do this
> is in place in all recent versions, since it has been use for grid
> computing.
If you mean what is described in the "GRID" examples, then I am already
doing it that way. The calculations would take far too long otherwise and
exceed the wall-time. But it doesn't help with only a single central disk
system. Also, when doing such separation of irrep it repeats the initial
wavefunction calculation (I think that is what it does at first) for each
irrep calculation or group of irrep, which in the case of large systems
becomes very time consuming and I've noticed can take longer than to
calculate a single irrep afterwards
Of course, I can always run on another system with distributed disks I just
have less resources there and these phonon calculation take a lot of time to
run.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20110802/e4f071ef/attachment.html>
More information about the users
mailing list