[Dune] Repartitioning/dynamic load balancing in Dune

Timo Koch timo.koch at iws.uni-stuttgart.de
Tue May 19 12:12:44 CEST 2020


Thank you Andreas and Oliver!

I will have a look at the examples and let you know if I experience any problems.

Timo
-- 
_________________________________________________

Timo Koch                                      phone: +49 711 685 64676
IWS, Universität Stuttgart                  fax:   +49 711 685 60430
Pfaffenwaldring 61         email: timo.koch at iws.uni-stuttgart.de
D-70569 Stuttgart             url: www.iws.uni-stuttgart.de/en/lh2/
_________________________________________________

> On 14. May 2020, at 10:25, Dedner, Andreas <A.S.Dedner at warwick.ac.uk> wrote:
> 
> 
> > 3) Does anyone have accessible/shareable application or testing code that uses dynamic load balancing via the dune grid interface? Ideally not using ParMETIS but e.g. Scotch or something else.
> 
> I forgot to mention: in dune-alugrid/examples/loadbalance
> there are some example of how to implement a loadbalancer from the outside and one example uses Zoltan - which might work for you.
> Andreas
> From: Dune <dune-bounces at lists.dune-project.org> on behalf of Oliver Sander <oliver.sander at tu-dresden.de>
> Sent: 14 May 2020 09:10
> To: dune at lists.dune-project.org <dune at lists.dune-project.org>
> Subject: Re: [Dune] Repartitioning/dynamic load balancing in Dune
>  
> Hi Timo,
> 
> > 1) Is the way implemented in the loadbalance test in dune-grid the way to currently do dynamic load balancing and will it work for dune-uggrid and dune-alugrid? Possible restrictions?
> 
> no, that is only for UGGrid.
> 
> As Andreas said, UGGrid wants a vector of element target ranks, and ALUGrid uses a call-back.
> So currently your code has to target one grid or the other, specifically.
> 
> > 2) What is the rational behind the new “repartition” interface and which problems will it solve?
> 
> I don't remember the discussion, but I suppose the aim was to unify the two interface mentioned above.
> 
> Indeed, I have not strong preference for the UGGrid interface at all -- it just seemed easiest
> to me at the time.  If we reach a consensus to use something else I'll by happy to change it.
> 
> > 3) Does anyone have accessible/shareable application or testing code that uses dynamic load balancing via the dune grid interface? Ideally not using ParMETIS but e.g. Scotch or something else.
> 
> You don't need any particular partitioner at all for UGGrid load-balancing.
> Just get a vector of desired target ranks for each element somehow,
> and give that to the loadBalance method.
> 
> That being said, that part of UGGrid has seen little testing, and you may encounter bugs...
> 
> In particular, I suspect that even UG posed some restrictions on how elements can be
> distributed, but that is not documented anywhere.  Peter is most likely to know anything
> about this.
> 
> Best,
> Oliver
> 
> > 
> > Already thanks for any answers in advance!
> > 
> > Timo 
> > 
> > 
> > -- 
> > _________________________________________________
> > 
> > Timo Koch                                      phone: +49 711 685 64676
> > IWS, Universität Stuttgart                  fax:   +49 711 685 60430
> > Pfaffenwaldring 61         email: timo.koch at iws.uni-stuttgart.de <mailto:timo.koch at iws.uni-stuttgart.de <mailto:timo.koch at iws.uni-stuttgart.de>>
> > D-70569 Stuttgart             url: www.iws.uni-stuttgart.de/en/lh2/ <http://www.iws.uni-stuttgart.de/en/lh2/> <http://www.iws.uni-stuttgart.de/en/lh2/ <http://www.iws.uni-stuttgart.de/en/lh2/>>
> > _________________________________________________
> > 
> > 
> > _______________________________________________
> > Dune mailing list
> > Dune at lists.dune-project.org
> > https://lists.dune-project.org/mailman/listinfo/dune <https://lists.dune-project.org/mailman/listinfo/dune>
> > 
> 
> _______________________________________________
> Dune mailing list
> Dune at lists.dune-project.org
> https://lists.dune-project.org/mailman/listinfo/dune

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20200519/24f86b1c/attachment.htm>


More information about the Dune mailing list