[CPMD-list] Parallelization on k-points

[Juerg Hutter]

Hi

in general a parallelisation over k-points is not
too difficult. However, in CPMD this would require
some work. Especially the FFT parallelisation would
have to be generalised.
Using Fortran90 this could be done using modular
programming and then a combined k-Point/G-vector
parallelisation would be possible.

Juerg

----------------------------------------------------------
Juerg Hutter                   Phone : ++41 1 635 4491
Physical Chemistry Institute   FAX   : ++41 1 635 6838
University of Zurich           E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------


On Sat, 5 Jul 2003 Ari.P.Seitsonen at iki.fi wrote:

>
> Dear Audrius,
>
> > We have a new cluster here in Basel, and I began running big jobs of
> > Ag(111) surface. I noticed that each k-point is done in serial. Would it
> > make a difference to do each of, let's say, 8 different k-points on a
> > different node? Is it difficult to implement? Would there be any time
> > economy or not?
>
> The k point parallellisation is/would be indeed _very_ efficient as
> the k points are very independent. It wouldn't be too much of work, I
> believe (I did it once in a couple of days, yet on an smaller code,
> once one knows the code it probably wouldn't take even that long. One
> important issue is of course I/O, whether to write the restart files
> in parallel, only on a single node etc).
>
>     Exhausted Greetings from again-sunny Zurich,
>
>        apsi
>
> --
> -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-
>  Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
>  Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35
>  Anschrift: Physikalisch Chemisches Institut (PCI), Universität Zürich (UniZh)
>  Indirizzo: Winterthurerstraße 190, CH-8057 Zürich
>  Address:   Schweiz / Svizzera / Suisse / Svizra / Switzerland
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://www.cpmd.org/mailman/listinfo/cpmd-list
>