Excerpts from Jimmy Shao's message of Fri Apr 30 13:31:23 -0500 2010: > Hello, > > This is Jimmy. > I am performing some iperf throughput test using 10 G and 1 G card. > > My set up as following: > > Server with 10G NIC---- (10G interface-SW ---1G interface)-----Client with 1G > NIC > > The problem: > > The iperf test (TCP throughput) result only shows the throughput only can > reach > up to 500M, and when using IXIA I can get up to 1G. > Is there any limitation in iperf for throughput test? > How many traffic (throughput) iperf can generate? > BTW, on switch I also turn off the" flow control" function. > > Any input and comments are welcome.
Hi Jimmy, I'd expect you to be able to easily get 990 Mb/s with this configuration assuming reasonable hardware for the test PCs (eg 1GHz or better). How long do you let the TCP test run? The autotuning (assuming you're running a newish Linux) can take a little bit to set up, so I'd suggest running for a few minutes rather than say just 30 seconds. Is there any cross traffic in your setup or is this the only traffic? What operating system and version is Iperf running on? Which version of Iperf? Single threaded or multi threaded? I haven't looked at Ixia's product line in quite some time, but last time I did they may have sent traffic that had TCP set in the protocol field of the IP header, but it didn't actually implement the TCP machinery. In other words it's blasting bits that say they're TCP but none of the TCP dynamics apply. It's possible that newer Ixia gear actually runs a TCP machine, but I kind of doubt it. Jon -- Jon M. Dugan <[email protected]> ESnet Network Engineering Group Lawrence Berkeley National Laboratory ------------------------------------------------------------------------------ _______________________________________________ Iperf-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/iperf-users
