DAT participants, This summarizes some of the issues that seem to be underlying the nameservice and provider registration discussions. Rajesh Sankaran wrote ...
Excellent summary. My comments are interspersed. ... I thought this was a settled issue. The rationale is: a) Consumer must not have to deal with providers, or...
Excuse my potential ignorance, but I have a question pertaining to the architecture pictures on the open source ib page: The architecture pictures all show...
I am scheduling extra meeting for Mondays starting 5/6/02 from 1:00-3:00pm so we can resolve Name Service and Registration issues ASAP! All other DAT issues...
It seems to me that the effort needed to define a common verbs API that the HCA vendors all agree to and implement is probably as big an effort as defining &...
Dat Collaborative members: Here is the conceptual proposal for the generic/detailed status codes that Sri briefly introduced during the face-to-face. We will...
The functionality of kDAPL may be sufficient for implementing SRP, but SDP and IPoIB require the UD transport, which is not supported in DAT 1.0. This...
Even though theoretially possible, the kDAPL spec. draft as it stands has multiple issues that needs to be resolved before one can implement SDP over it. Some...
While the SDP transport is implemented over RC, it uses IPoIB to perform the IP to GID/LID address resolution for connection setup, which engages UD. Alfred...
If the intended usage for kDAPL (not uDAPL), is kernel level ULPS like SRP, SDP, IPoIB, whatever, then I assert that what the IB HCA community should do is to...
... It is easier to be silicon-independent (verbs-like channel interface) than to be both OS-agnostic and transport-independent (kDAPL). So an SDP implementer...
As per the IB specification all HCA vendors have to provide verbs interface. DAT also needs a standard verbs interface, otherwise it writes to several...
Thanx for the responses everyone. Another thing I didn't consider is that IPoverIB and SDP are really IB-specific. I'm not sure about SRP? But things like...
Steve, ... (hi ... to use ... a CA or ... This is true. Applications which need RDMA abstraction will use something like DAT. IB specific applications (SDP,...
... Som, SDP and SRP are not IB specific applications. They are protocols that aim to be transport agnostic. (An iWARP SDP wire protocol will be defined.) By...
David Ford
dford@...
May 2, 2002 8:41 pm
1031
Dave, As it stands today, SDP And SRP are defined on InfiniBand. In fact, SDP is supplement to InfiniBand Architecture specification volume1 (Annex A4). So...
... Dave, isn't the path migration stuff in SDP IB-specific? So there will be a different SDP specification for iWARP SDP... It's sort of a nit, but I think...
... The first goal was to be compatible with server clusters where the clustering is administered in an application independent fashion. In our experience, ...
I'd like to pop up a level here. Yes, there are existing implementations of SDP, SRP and IPoIB that are coded to various verbs layers. That's what existed at...
David Ford
dford@...
May 2, 2002 10:12 pm
1036
... All excellent questions. And the answer is, we don't know. Or to phrase it more positively we don't care. The API defines what the application requires,...
... Short party-line answer: Buffer advertising is a ULP issue. Use of RMR Contexts to access memory is allowed *ONLY* to the extent that the upper layer...
... That's too early to say now. We can comment on this only after we see DAPL, Access Layer, Verbs, CAD layer implementations and performing a a comparison. ...
I would like to add for duplicated connection: if Consumer specifies the same Protection Zone/Domain (PZ) for the duplicated connection then these connections...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
