[openib-general] Re: SA MultiPathRecord v.PathRecord

[Hal Rosenstock]

On Thu, 2006-05-04 at 13:07, Sean Hefty wrote:
> Moving discussion to list.
> 
> Hal Rosenstock wrote:
> > MPR does allow for better selection of paths for APM though.
> 
> Beyond adding the Independence Selector, does MPR add anything else?

Yes, there is 1.2 erratum which extends SA MPR by adding scope for both
the S/DGID which allows for choosing GIDs from same HCA or from same
system (based on SystemImageGUID) as well as explicit which is the way
it is in the 1.2 spec.

> I assume that the Independence Selector works when paths between multiple GIDs 
> are requested.

Yes but with scope set to more than just explicit, this can be between
individual GIDs (and the SM calculates the rest so the end client
doesn't need to figure out the other GIDs).

> Have you thought about how to expose this capability to a client 
> in such a way that they can make use of it?

No; not yet.

> To me, this feature seems most useful for all-to-all type connections, but would 
> require some sort of coordination between connecting end-points in order to have 
> fault independent connections between different nodes.  E.g. the connection from 
> A to B is independent from the connection between C and D.

I don't understand what end node coordination you are referring to here
in terms of this.

> Also, for large fabrics, even MPR seems limited by the GID counts being 8-bits.

If there was a half (one of source and dest wildcarded) and full world
(both source and dest wildcarded) capability like PathRecord that would
not be a limit.

-- Hal

> - Sean