[CP2K:7670] Example for dynamic load imbalances

[Iain Bethune]

Hi Martin,

In CP2K, load imbalance typically occurs because of some spatial properties in the system being modelled.  For example, systems with an interface between two materials, or a surface or cluster in vacuum.  I attach an input for a cluster of water molecules which was used in this paper: https://cug.org/5-publications/proceedings_attendee_lists/CUG10CD/pages/1-program/final_program/CUG10_Proceedings/pages/authors/16-18Thursday/17B-bethune.pdf  However, if run on a large number of MPI processes, even the standard water box benchmarks (in cp2k/tests/QS/benchmark) will show some imbalance.

The load balancing algorithm is run whenever the geometry changes, so it will adjust the task decomposition (which processes handle which gaussian products), and the reordering of the realspace grids as the calculation goes on, for example during MD.  Since a few months ago, the algorithm is quite memory efficient, so there is no need to turn it off (although the FORCE_EVAL / DFT / MGRID / SKIP_LOAD_BALANCE_DISTRIBUTED flag may still be used).

- Iain

--
-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: W216.inp
Type: application/octet-stream
Size: 23091 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160414/ee1e3758/attachment.obj>
-------------- next part --------------


Iain Bethune
Project Manager, EPCC

Email: ibet... at epcc.ed.ac.uk
Twitter: @IainBethune @PrimeGrid @CP2Kproject
Web: http://www2.epcc.ed.ac.uk/~ibethune
Tel/Fax: +44 (0)131 650 5201/6555
Mob: +44 (0)7598317015
Addr: 2404 JCMB, The King's Buildings, Peter Guthrie Tait Road, Edinburgh, EH9 3FD

> On 13 Apr 2016, at 09:53, Martin K?ttler <martin.... at gmail.com> wrote:
> 
> Hi,
> 
> I'm interested in the load balance behavior of CP2K. I've done some measurements already, but I mostly found inputs that lead to small and mostly static load imbalance. I am aware of the built-in load balancer, and I turned it off for my measurements. Can somebody point me to imput files that exhibit bigger imbalance, and maybe even changing imbalance over time? I'm using CP2K mostly as a benchmark, so unfortunately my knowledge of the actual physics is very limited.
> 
> Thanks!
> Martin
> 
> -- 
> You received this message because you are subscribed to the Google Groups "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com.
> To post to this group, send email to cp... at googlegroups.com.
> Visit this group at https://groups.google.com/group/cp2k.
> For more options, visit https://groups.google.com/d/optout.