[mpas-developers] Multiple Blocks per Processor

[Doug Jacobsen]

Hi All,

In exploring potential performance improvements for MPAS ocean a discussion
on the option of multiple blocks per process came up. Several members of the
mpas-ocean group are very interested in being able to use multiple blocks
per processor. This capability might enable addition performance
enhancements to MPAS cores that is not currently available. In order for
this to be implemented current issues and potential roadblocks need to be
identified. I do not have a significant amount of experience with the block
decomposition in MPAS aside from what (k/p)metis does. Having multiple
blocks per process will facilitate implementation of OpenMP directives and
off-loading to GPUs/accelerators.

If anyone would like to assist in this project, with actual implementation,
or suggestions/comments/discussion it would be very appreciated. Also, if
anyone has any reasons why this would not be a good idea that would also be
appreciated.

Thanks!

Doug
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20110928/a2f04476/attachment.html