[Mpi3-tools] URGENT: feedback from first reading of MPIR

John DelSignore John.DelSignore at roguewave.com
Thu Sep 16 11:23:31 CDT 2010


Jeff Squyres wrote:
> The reading went generally well.  There were a few (legitimate) nit-picks from the Forum that I need to fix tonight and re-present tomorrow.  Doing this will still allow us to meet the requirement of "first formal reading" at this Stuttgart Forum meeting -- paving the way for the first vote next meeting, etc.
> 
> Here's a list of the nits:
> 
> - add version number for MPIR document
> - It should to be mentioned explicitely that the document is _NOT_ part of MPI 3 and does not have to be implemented for MPI 3 compliance
> - Is "IP address" only IPv4 ? 

No, it doesn't have to be an IPv4 address. There is nothing that I know of in MPIR that excludes IPv6. The "host_name" field of the procdesc structure is a string pointer, so the string can resolve to an IPv4 or IPv6 address. And as it says in the document, on BProc systems, "host_name" is a BProc node number so the string is also open to interpretations other than just IP address.

Cheers, John D.


> - Message Queue has not been mentioned as a separate document
> - MPI_Comm_join is missing on the list of unsupported functionality
> - Fix Link to page number in bibliography (It is irritating as they _all_ point to page 1)
> - MPIR_debug_state could be set by a compiler default 
> - My name (Marc-Andre) as Marc-Andr\'{e} 
> - inconsistent use of names in chapter 2 (MPI APA) vs. rest of document
> 
> I will be fixing all of these in the next hour or two.  The only real question is "IP address" -- do we only mean IPv4?  Or is IPv6 ok?  I'll read the text closely to see what we intended.  If anyone has an opinion, please chime in ASAP.  Thanks.
> 



More information about the mpiwg-tools mailing list