On Sep 22, 2010, at 8:04 AM, Jens Domke wrote:
> I already tried this parameter, but I don't see any improvements in the benchmarks. Additionally while doing further investigations into the opensm I didn't see the QP requests for other LIDs than the base LIDs.
I'm afraid that I'm not an InfiniBand vendor anymore, so I haven't looked at this code in a long, long time. Can an IB vendor comment here?
One thing to be aware of is that LMC may or may not have much effect on benchmarks. Remember that LMC is NOT a bandwidth multiplier -- you still have the same links (and total bandwidth) out of each port. LMC>0 only (potentially) sprays packets across more routes on your network, which *may* (or may *not*) have effects regarding congestion. If you're running a single app on an otherwise-quiet IB network where all distances are the same (or short enough not to matter), you may not see much difference.
That being said, if you were watching OMPI's behavior with regard to LIDs and only saw the base LIDs go by, then perhaps something is wrong. Shrug.
For corporate legal information go to: