[Trilinos-Users] Zoltan repartition question

[Devine, Karen D]

The LB_APPROACH parameter (REPARTITION) is used only by Zoltan's graph and
hypergraph partitioning methods, not by the geometric methods (RCB, RIB,
HSFC).  RCB makes cuts perpendicular to the longest axis of the bounding
box, independent of the initial partition.  And while Zoltan's REMAP flag
renumbers the resulting parts to improve overlap with the initial
partition, for this particular example, it does not help.

You can try using the parameter RCB_SET_DIRECTIONS to tell RCB which cut
directions to use first.  It should help for this particular example.  See
documentation at http://www.cs.sandia.gov/zoltan/ug_html/ug_alg_rcb.html

Karen


On 10/14/11 2:59 AM, "Biddiscombe, John A." <biddisco at cscs.ch> wrote:

>Generating a synthetic dataset on two cores, and then running a zoltan
>partition, I expected that using the Œrepartition¹ option that the data
>would probably be redistributed in almost the exact same location as it
>started in.
> 
>However the original (left) and repartitioned (right) are divided along a
>different axis.
> 
>Should I expect this behaviour? Can I supply any hints to Zoltan to
>minimize the transfer of data? (I presume I¹ve missed some flag somewhere)
> 
>I use ...
> 
>Zoltan_Set_Param(zz,
>"LB_APPROACH", "REPARTITION");
> 
>Zoltan_Set_Param(zz,
>"LB_METHOD",   "RCB");
> 
>Thanks
> 
>JB
> 
> 
>
> 
> 
>-- 
>John Biddiscombe,                            email:biddisco @ cscs.ch
>http://www.cscs.ch/
>CSCS, Swiss National Supercomputing Centre  | Tel:  +41 (91) 610.82.07
>Via Cantonale, 6928 Manno, Switzerland      | Fax:  +41 (91) 610.82.82
> 
> 
>
>
>
>_______________________________________________
>Trilinos-Users mailing list
>Trilinos-Users at software.sandia.gov
>http://software.sandia.gov/mailman/listinfo/trilinos-users