[openib-general] [PATCH 0/4] SA path record caching
Hal Rosenstock
halr at voltaire.com
Tue Jan 31 11:02:27 PST 2006
On Tue, 2006-01-31 at 14:01, Sean Hefty wrote:
> Hal Rosenstock wrote:
> >>I see where a MultiPathRecord contains a path independence field. But since it
> >>only provides a single value for all other PathRecord fields for all SGID/DGID
> >>pairs, it seems less flexible than PathRecords.
> >
> > It depends on whether those components are wildcarded or not.
>
> I was referring to the query response.
The query response is PathRecords not MultiPathRecords.
>
> >>It's also missing LID information,
> >
> > Yes but is LID needed in the query ?
>
> It's needed for connection establishment and QP modification.
but in the response, right ?
> >>and limits the number of DGIDs to 255.
> >
> > Actually the number if lower as I think it is limited by the remaining
> > bits in the component mask.
>
> For a fabric with 1000 or more nodes, the MultiPathRecord only returns paths to
> 255 of the nodes, and then only if their path information matches. (This is
> part of why I'm confused about how to actually use a MultiPathRecord.)
You can wildcard the DGID (and DGIDCount).
> From what I can tell, a MultiPathRecord does not eliminate the need to obtain
> PathRecords,
It's a better way to get PathRecords.
> and only provides a 2-bit path independence field that's always set
> to 1. It seems like I'm missing something here.
Aside from the independence field, there's 2 new fields" SGID/DGIDScope
to be used for explicit/node/system GID indication.
-- Hal
More information about the general
mailing list