interoperability with RDMAC and IETF

"Jim Pinkerton" <[email protected]> Tue, 9 Nov 2004 13:42:42 -0800
Newsgroups gmane.ietf.rddp
Message-ID <E6564B8F86852D46A4E98C485FB33B8F0ADF8FF7@WIN-MSG-10.wingroup.windeploy.ntdev.microsoft.com>
A quick update on interoperating between RDMAC and IETF RDMAP/DDP/MPA
wire protocols. Wanted to get a quick summary of a possible approach out
to see if the work group felt this was a reasonable approach.

 

Post meeting, over several beers last night, John Carrier and I think we
have a solution that defines a 3x3 interoperability matrix where 7 out
of the 9 cells allow interoperability (the other 2 must abort because
markers were disabled on one side but not the other). We're going to
write up a proposal soon. It entails minor changes to IETF MPA
(primarily change the version number of the initial negotiation to one),
and also defines the new capabilities that would need to be present in
an IETF RNIC if the implementer wants to be downward compatible (primary
issue is being able to tell the RNIC for a specific QP to generate
DDP/RDMAP packets with version number = 0). 

 

It entails no changes to the RDMAC specifications, but does require
RDMAC RNIC ULPs to communicate in streaming mode prior to transition to
RDMA mode, and that the format of the data be the MPA initial
negotiation (REQ/REP) frames defined within the IETF, but using a
version number of zero.

 

David, does this high level description of the approach seem reasonable?

 

 

Jim

_______________________________________________
rddp mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/rddp