[Trilinos-Users] zoltan particle ghost regions

[Devine, Karen D]

You do not have to use lists provided by Zoltan in Zoltan_Migrate.  You
can build your own lists, as long as your arguments to Zoltan_Migrate are
consistent with Zoltan's lists and the callback functions
(ZOLTAN_PACK_MULTI_FN, ZOLTAN_UNPACK_MULTI_FN, etc.) can interpret the IDs
correctly.

Karen


On 5/23/11 9:01 AM, "Biddiscombe, John A." <biddisco at cscs.ch> wrote:

>Lee Ann,
>
>I got back to this and particle repartitioning is working nicely -  I was
>pondering the ghost cell sending.
>I looked again at the migration example page and wondered ...do I need to
>bother with Zoltan_Invert_Lists
>it seems to me that I can just make up my own lists (the zoltan_rcb_box
>works fine and helps) of Ids and destinations, package them up in a call
>to zoltan_migrate and do everything 'custom'.
>
>Can I call Zoltan_migrate with all my own data pointers? is there any
>reason why I have to use lists zoltan has created? I'm guessing that I
>can do what I want providing I package up everything using the callbacks,
>but I may find out the hard way that I can't - hence I'm asking
>beforehand ...
>
>thanks
>
>JB
>
>-----Original Message-----
>From: Riesen, Lee Ann [mailto:lriesen at sandia.gov]
>Sent: 03 May 2011 20:36
>To: Biddiscombe, John A.; trilinos-users at software.sandia.gov
>Subject: RE: zoltan particle ghost regions
>
>I should also mention that if you are partitioning with axis-aligned
>spatial regions (the RCB method) you can use Zoltan_RCB_Box to obtain the
>boundaries of the region, which may help you determine how close your
>points are to a region.
>---
>Lee Ann Riesen
>Sandia National Laboratories, Albuquerque, NM, USA
>Department 1426, Scalable Algorithms, Computing Research Center
>505-844-2059
>________________________________________
>From: Riesen, Lee Ann
>Sent: Tuesday, May 03, 2011 12:31 PM
>To: Biddiscombe, John A.; trilinos-users at software.sandia.gov
>Subject: RE: zoltan particle ghost regions
>
>Hello John.  Zoltan does not explicitly support any level of ghost
>points.  It simply maps each point being partitioned to a single part
>number.  When the application migrates the points to the new part, it can
>also send neighbors that are not assigned to the new part as ghost points.
>
>If you are using Zoltan to migrate the points, you can include in your
>callback functions the movement of the ghost points.  That is what is
>being described in the example you referenced.
>
>I hope this helps.
>
>Lee Ann
>---
>Lee Ann Riesen
>Sandia National Laboratories, Albuquerque, NM, USA
>Department 1426, Scalable Algorithms, Computing Research Center
>505-844-2059
>________________________________________
>From: trilinos-users-bounces at software.sandia.gov
>[trilinos-users-bounces at software.sandia.gov] on behalf of Biddiscombe,
>John A. [biddisco at cscs.ch]
>Sent: Monday, May 02, 2011 3:48 AM
>To: trilinos-users at software.sandia.gov
>Subject: [Trilinos-Users] zoltan particle ghost regions
>
>If using zoltan to redistribute particles between processes. Is it
>possible to ask for all points within x units of a boundary to be
>duplicated as ghost points and copied to all overlapping processes.
>
>I had a quick look here which gives me some clues, but if there's an
>example already out there, please let me know.
>http://www.cs.sandia.gov/zoltan/ug_html/ug_examples_mig.html
>
>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
>
>
>
>_______________________________________________
>Trilinos-Users mailing list
>Trilinos-Users at software.sandia.gov
>http://software.sandia.gov/mailman/listinfo/trilinos-users