> > >> I'll be posing some data soon > >> (4 us latency and 5000 Mbits/sec using OpenMPI and netpipe) > > [...] > >> But as the core counts on the motherboards get higher, IB > gets more > >> attractive. > > > > One wonders why you're still using single-core benchmarks on > > multi-core systems... > > what's the alternative? single-threaded lat and bw are still > fundamental properties of the interconnect, just less of the > whole story than before.
I have moved to test latency with multi-processes, each running on a single core in a multi core environment. Good interconnects will show almost the same latency for each process, when the process are executed in parallel. With other solutions, the latency will increase with the number of process. Gilad. _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
