On Dec 12, 2013, at 4:31 PM, "Prindeville, Philip" <philip.prindeville_at_[hidden]> wrote:
> I think I understand.
> Ill pull a copy of trunk and dig around in there.
> Is there a reason that the code cant be gated by conditional compilation flags or detect the environment at run-time?
I'm not sure what you're asking -- we have a bunch of code that is gated on #if's, and we have a lot of auto-detection at run-time. That's kinda OMPI's mantra: look around the system at run time and use what it finds. Meaning: I think what you're suggesting is already fairly common practice in this community.
Are you thinking of modifying the openib BTL to do more things with SCTP as a lower-layer transport for iWARP (under verbs)? To do so, you'll need to go outside the verbs API -- is that correct?
I'll warn you that the openib BTL is fairly complex. :-) Indeed, if you want an iWARP+SCTP-specific transport, it *might* be worthwhile to do your own BTL...? You might even be able to use openib as a starting point, strip out everything that iWARP doesn't need (i.e., all the IB/RoCE-specific stuff), have something quite a bit simpler than openib, and then add your SCTP stuff on that...? Just an idle thought.
Note that if you're going to contribute code back to Open MPI, we'll need a signed contributor's agreement (see http://www.open-mpi.org/community/contribute/).
> Is there anything in the way of a set of verification tests?
We typically just use lots of standard MPI tests and benchmarks.
> And whats the provenance of the SCTP code thats in trunk?
It was originally written by U. British Columbia. It doesn't use iWARP at all. It also hasn't been maintained in quite a while -- I can't honestly say what its current state is (which is one of the reasons it was removed from the v1.7 release branch).
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/