[Carpet] proc distribution

[Erik Schnetter]

On Aug 22, 2006, at 08:05:58, Michael Koppitz wrote:

> Hi,
>
> just since I am (kind of) new allow me this question
> people probably asked several times:
>
> Would it be sensible to distribute the processors
> more cleverly? is it true that currently each proc
> has its share from each refinement-level?
> especially with the new architectures wouldn't it
> reduce time spend in communication if we would
> give each proc (or better still: node) a cube of points
> on one level only?

The standard Berger-Oliger time evolution algorithm evolves the  
levels sequentially.  That means that distributing each level onto  
all processors is best, and this is what Carpet is doing.

One could distribute the processors on level 1 and on level 2 such  
that the inter-level communication for prolongation and restriction  
is optimised.  This is currently not the case.

When one uses global time stepping for all or for some of the levels,  
then one could modify the Berger-Oliger algorithm to evolve these  
levels in parallel.  In this case, such a different distribution  
would improve performance.

-erik

-- 
Erik Schnetter <schnetter at cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : /archives/developers/attachments/20060822/3ef50589/attachment.pgp