[Trilinos-Users] [EXTERNAL] 2 zoltan questions

Biddiscombe, John A. biddisco at cscs.ch
Wed Dec 19 14:25:31 MST 2012


In the unpack callback, there are no guarantees about the Ids being monotonically increasing.  I suspect you are getting this lucky behavior because your Ids are nicely ordered within processors to begin with.  

OK understood. I suspect I'm just getting lucky then. However, the monotonic requirement won't be a show stopper it just makes my insertion check simpler. I'm using a std::map, but the key is the GID and the value is the offset, if I do a lower/upper bound on the GID and check the offset (diff between GID and LID) of the last point and it's the same as the point I'm inserting, then I don't need to store the new point, just assume the offset is the same. Saving me a lot of memory for the moment.

For purely on-processor data, you can efficiently map global Ids to local indices with a hash table or std::map.  For mapping global Ids to remote local ids, you can use the Zoltan distributed data directory.  If you decide to use the data directory, I am happy to provide instructions.

Thanks. I'll look at this.

Yes, the interface to Zoltan2 is much different from Zoltan.  In Zoltan2, users will instantiate an input adapter that describes their problem; input adapters currently supported describe meshes, matrices, coordinates, identifiers, and graphs.  The benefit is that users who have, say, a mesh, no longer will have to convert their mesh to graphs or hypergraphs;
Zoltan2 will do the appropriate conversion based on mesh adjacencies.
While we plan to create a Zoltan1-input adapter to allow easier conversion to Zoltan2, this adapter is not yet available.  Which partitioning methods are you using?

OK. I hope main code remains unchanged. (I do load-balance points, with auto-migrate, then locate ghost points or cells I want to exchange, do invert lists and manual migrate - I presume (hope) this basic sequence will stay the same even if the mechanism by which I invoke it changes. I'm using plain RCB at the moment but eventually will want to experiment with others one day :) 



More information about the Trilinos-Users mailing list