[mpas-developers] Design Document: Non-blocking halo exchanges

Doug Jacobsen jacobsen.douglas at gmail.com
Thu Feb 21 09:43:55 MST 2013


Hello Everyone,

I have been working on implementing a non-blocking version of our halo
exchange routines. In order to get some input on what people think about
them, I put together a design document.

Although the design document has a proposed benefit that applies directly
to cores that support multiple blocks, there is a potential benefit for
cores that don't support multiple blocks too.

Non-multiple block cores could break up their halo exchanges to group the
initialization of all halo exchanges for all fields, followed by the local
copies for all fields, and finally the finalization of all halo exchanges
for all fields. This could allow more computation between the
initialization of the MPI_Isend/MPI_Irecvs pairs and the MPI_waits used
before unpacking the buffers.

Anyway, please let me know if you have any comments or questions. I've also
committed this to the repository.

Thanks,
Doug
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20130221/68490371/attachment-0001.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: non_blocking_halo_exchanges.pdf
Type: application/pdf
Size: 152685 bytes
Desc: not available
Url : http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20130221/68490371/attachment-0001.pdf 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: non_blocking_halo_exchanges.tex
Type: application/x-tex
Size: 8375 bytes
Desc: not available
Url : http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20130221/68490371/attachment-0001.tex 


More information about the mpas-developers mailing list