Hi Al, Thanks for doing those tests, I think they clearly demonstrate the importance of doing the right routing when running with LMC>0, actually it is important to do correct routing anytime ;-)
We can also see that Open MPI manage to get significant performance improvement (faster by up to 38%) when utilizing 4 routs (with PL patch) and mange to maintain scalability of this feature for almost 1000 ranks. We are thinking on additional ways to improve Open MPI utilization of multi routs so we can even get better performance in the future. Yiftah > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:general- > [EMAIL PROTECTED] On Behalf Of Hal Rosenstock > Sent: Tuesday, June 17, 2008 19:43 > To: [EMAIL PROTECTED] > Cc: Yevgeny Kliteynik; [EMAIL PROTECTED]; OpenIB > Subject: [ofa-general] Re: [PATCH] opensm: preserve base lid routes > > On Tue, 2008-06-17 at 09:25 -0700, Al Chu wrote: > > Sorry, the numbers for the mvapich/openmpi numbers are in units of > > time. > > > I think microseconds. So lower is better. > > That's what wasn't clear. > > -- Hal > > _______________________________________________ > general mailing list > [email protected] > http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib- > general _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
