From: Garcia M. M. <mar...@iw...> - 2004-12-10 18:48:28
|
My name is Marc Garcia and I am doing my Diploma-Thesis in the Forschungszentrum Karlsruhe in the field of Networking in the Gridka project (www.gridka.de). Some days ago I sent you an Email explaining you some strange results within our Gigabit ethernet network and your software (iperf) but we received no answere. Using a Tyan motherboard chipset E7501 dual Pentium, Intel Corp. 82546EB Gigabit Ethernet Controller and iperf version 2.0.1 (lastest one) we obtained strange results in a back to back UDP test, iperf gave us around 500Mbps with a high packet loss but under the same conditions, the behaviour of an TCP test was stable and close to 1Gbps. We tried the same test in other equipment (IBM) and both test succeeded with preety and stable results; no errors and both protocols were close to 1Gbps. We thought it was a hardware problem and after analising the device driver of the first group of equipments the problems still persisted; far from the good IBM results. Connecting the nodes back to back or via a switch did not change the results at all. We also observed that when using -l parameter and a bigger value than the default one (1700 and upper), it worked perfectly; no errors, very good throughput... As I told you in my last email, for us it didn't make sense (and still it doesn't) because the MTU of the GEthernet is set to its default (1500 bytes) so the TCP/IP must fragment the packet before sending it and then the 2 second packed won't be fullfilled so in principle the performance should be lower but it was not. On the other hand, we have decided to install the previous version of iperf (1.7.0) and it works simply great. I attach you the results we gathered with the hope it may clarify. We intended to use Iperf till version 1.7.0 as a network quality measurement tool. Since result of version 2.0.1 is so different, could you please explain. Thanks a lot in advance for your time, Best Regards, Marc Garcia Marti mar...@iw... Forschungszentrum Karlsruhe Tel.: +497247828601 Institute for Scientific Computing / Inst. f. Wissenschaftliches Rechnen Hermann-von-Helmholtz-Platz 1, 76344 Eggenstein-Leopoldshafen, Germany |