[Trilinos-Users] The right matrix in Epetra for the adjacency matrix of a power-law graph

[Boman, Erik G]

Sorry, the code from our SC13 paper is not in Zoltan or Zoltan2. Still, 2-dimensional distributions are important if you do parallel computations on power-law graphs.

Please tell your sponsoring funding agency/company if this is something you need in Trilinos ;)

Erik


On Mar 17, 2015, at 8:06 AM, Zheng Da <zhengda1936 at gmail.com<mailto:zhengda1936 at gmail.com>> wrote:

Thank you for your reply, Erik.
I see you have a paper on sparse matrix vector multiplication, published in SuperComputing 13.
I wonder if I can access the code?

Thanks,
Da

On Mon, Mar 9, 2015 at 1:30 PM, Erik Boman <egboman at sandia.gov<mailto:egboman at sandia.gov>> wrote:
Da,

Just use Epetra_CrsMatrix. Power-law graphs do not have (small) dense blocks like some PDE problems do.

Erik

Zheng Da wrote:
Hello,

I just start to use trilinos. I want to find the best Epetra matrix
for the adjacency matrix of a power-law graph. I can use
Epetra_CrsMatrix to store the adjacency matrix, but I'm thinking of
having the adjacency matrix split into blocks in order to achieve
higher CPU cache hits.

It seems Epetra_VbrMatrix does something similar, but I can't find out
how the data in Epetra_VbrMatrix is organized. It seems that
Epetra_VbrMatrix stores each block as a dense matrix. If so,
Epetra_VbrMatrix won't be memory-efficient for a power-law graph even
if the graph is well partitioned/clustered.

Are there any other matrices I can use for the adjacency matrix in
Epetra or other packages?

Thanks,
Da
_______________________________________________
Trilinos-Users mailing list
Trilinos-Users at software.sandia.gov<mailto:Trilinos-Users at software.sandia.gov>
https://software.sandia.gov/mailman/listinfo/trilinos-users



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://software.sandia.gov/pipermail/trilinos-users/attachments/20150318/772996d3/attachment.html>