Or Gerlitz wrote: > Arlin, > > I see now that the uDAPL CMA provider code uses the MTU 1:1 as > returned by the SM in the path, so if the env is made of the Mellanox > PCI-X HCA there can be big BW drop, etc... we have discussed that. > > I wonder how are you overcoming this when running Intel MPI w. OFED 1.0?
We are not. The problem exists with the CMA provider and OFED 1.0. > > I understand in OFED 1.1 there is this tavor_quirk in both the cma and > the opensm, but i am not aware to any such hack in OFED 1.0. > > Also, i understand that OFED includes the uDAPL **SCM** provider, is > it really tested/supported? if yes, i don't think it needs to be. It > adds the overhead of one TCP connection per IB connection, creates two > codes bases to maintain, makes the CMA less tested, you named it. > > If its not tested/supported sure we must not provide it. > > If you agree would you approach the OFED maintainers to remove the SCM > provider from the udapl OFED 1.1 RPM? You are correct. There is no need to support SCM moving forward. -arlin _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
