[Wien] Wien2k with load-distributing systems

[Jorissen Kevin]

Machines-file!
Every line in the file (except special stuff like granularity, residue
...) corresponds to a set of k-points.  The first number is a weight -
it determines the relative size of this particular k-point set.  After
that, you write the machines to be used for this set of k-points.
*If you give just one processor for each set/line [eg.:
1:slowmachine
2:fastmachine]
, then you're doing 'k-point parallellization' (and the fast machine
gets twice as many k-points as the slow machine).
*If you write just one line  and specify more processors [eg. : 
1: fastmachine:2 slowmachine:1]
, then you do 'fine-grained' parallellization, i.e. 'real' parallel
computing (in this case, two processors of fastmachine and one processor
of slowmachine cooperate for each k-point).

*Of course you will often mix them.  Suppose you have two dual machines,
and more than one k-point, you can use :
1: dual1:2
1: dual2:2
Which means that the k-list will be split into two parts (k-point
parallellization), and each part will be calculated by 'fine grain
parallellization' by two processors of one dual machine.

Kevin.


-----Oorspronkelijk bericht-----
Van: Michael Frotscher [mailto:frotscher at chemie.uni-hamburg.de] 
Verzonden: woensdag 17 september 2003 9:53
Aan: wien at zeus.theochem.tuwien.ac.at
Onderwerp: Re: [Wien] Wien2k with load-distributing systems


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Wednesday 17 September 2003 03:53, Valerio Bellini wrote:
> If you are on a shared-memory system, you should not use the USEREMOTE

> variable when you make the 'siteconfig' setup. Instead you should use 
> some parallel environment command like poe, or mpiexec or 
> mpirun..depending on your computer.

I guess I'm better off using mpi - at least its installed and I can use
it. 
Using ssh is not such a good idea after all, as the ssh-forks are then 
one-by-one submitted to the batch system, which treats each of them as a

different batch job and executes them one after the other. This sort of 
defies the meaning of parallel processing.

I'll try to recompile WIEN using the shared memory architecture. As I
don't 
really get that section of the userguide: how do I tell WIEN to use the 
finegrained parallel instead of k-point parallelization?

Sincerely,
- -- 
Michael Frotscher
Institute of Inorganic and Applied Chemistry
University of Hamburg, Germany 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2-rc1-SuSE (GNU/Linux)

iD8DBQE/aBLM/f+kgY+d9bQRAoXcAJ91WKZdC/AeBynX8RoO3fH6IPpU/QCfbQgi
LJwub3QZQFkLiCSV98C2QJ8=
=TCjK
-----END PGP SIGNATURE-----


_______________________________________________
Wien mailing list
Wien at zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien