[gpaw-users] How does amount of K-points affect the performance of a run?

[Ask Hjorth Larsen]

Dear Vladislav

2012/7/9 Vladislav Ivanistsev <olunet at gmail.com>:
> Dear Ask,
> could you please specify, how could I force parallelization over k-points.
>
> In a script there is this row: kpts=(3, 4, 1) without specifying parallel
> options. Than in the output I see:
> Total number of cores used: 120
> Parallelization over k-points: 6
> Domain Decomposition: 2 x 2 x 5
> Diagonalizer layout: Serial LAPACK
>
> Does this mean, that not all (3*4) k-points are parallelized? How could I
> suppress domain decomposition?
>

First: Look in the text file to see how many k-points are in the
*irreducible* Brillouin zone, as those are the only ones that actually
take part in the computation.  Here you have 12 k-points which are
probably reduced by time reversal symmetry to 6, all of which are
parallelized.

Second: If for some reason you find yourself in a situation where this
is not the case, you can get it to parallelize over k-points in a
specific way by specifying the exact domain decomposition and band
parallelization, e.g. parallel={'domain': (2, 2, 1), 'band': 1}.  If
that does not work out to include all the CPUs, any k-points will be
distributed as evenly as possible among the CPUs not required to
fulfill the parallelization specification.  This may lead to some load
imbalance (maybe some CPUs will have 7 k-points while others have 6.
Make sure the load imbalance isn't too crazy, like 1:2)

Regards
Ask