[openfabrics-ewg] Getting local_sa cache support and multicast support into OFED 1.2

Michael S. Tsirkin mst at mellanox.co.il
Thu Jan 18 09:26:30 PST 2007


> Quoting Sean Hefty <mshefty at ichips.intel.com>:
> Subject: Re: Getting local_sa cache support and multicast support into OFED 1.2
> 
> > Sean should let us know when multicast kernel code can be put in OFED,
> > (I assume it is in multicast branch of his tree). Note that Voltaire seem
> > to want extensions to code, if so speak up, not a lot of time left for freeze.
> 
> The multicast code is in my multicast branch.  I plan on submitting a request to 
> pull in the multicast changes to ib_sa and ib_ipoib (and possibly rdma_cm) for 
> 2.6.21, so it can start to get wider testing.
> 
> Voltaire is asking for changes to the rdma_cm's use of the multicast code, 
> specifically, so that they can offload ipoib.  The actual code changes that they 
> require do not appear to be large.

OK, so one hopes it'll be ready by feature freeze.

> > For userspace, it was agreed that we will ship a release of librdmacm, correct?
> > So I don't understand how taking some devel branch in moves us forward.
> > But, we can live with it I guess. Sean needs to specify where (which branch)
> > to take librdmacm for OFED 1.2 from, and same for Arlin for DAPL.
> 
> Which branch depends on the users.  The most stable branch is likely the 
> rdma_ucm-abi3 branch, which matches up with what's in 2.6.20.  This does not 
> include multicast support, however.  The librdmacm master branch includes the 
> multicast support.

Hmm, this makes sense, and that's what we are building already.
Woody, why did you ask us to build from another branch?
What about DAPL? Arlin?

> Voltaire would like multicast support in OFED 1.2.  Intel would like multicast 
> support in a release, but I don't believe hitting OFED 1.2 is a strict 
> requirement for us.  Since Voltaire is requesting multicast support, my goal is 
> to try to get that into OFED.

But, we still ship a release of librdmacm, correct?
We are taking stuff from master but at feature freeze
it will become release 1.0 branch, and by code freeze
there will be a release 1.0 tag?

-- 
MST




More information about the ewg mailing list