1

I used iperf to measure the max throughput for TCP connection between two ubuntu machines. Both the machines were directly connected be means of a 1Gbps cable. When I performed iperf tests i obtained a bandwidth of 845 Mbps on an average. But when i tried reversing the direction (i.e changed server to client and viceversa) the results went down to 185 Mbps.These results are an average of ten readings and the max i was able to obtain was 370Mbps. I then decided to set the same value for parameters such as rmem_max, wmem_max, tcp_rmem, tcp_wmem but even after setting same parameters i got similiar results. i.e(845Mbps in one direction and 185Mbps in the other direction). I don't know what the possible cause could be. Is there something important that i have missed?

these are the results of netstat -in on both the systems:

Ifac MTU    Met   RX-ok   rx-err rx-drp rx-ovr tx-ok tx-err tx-drp tx-ovr flg 
eth0 1500    0     14976     0      0     0     8546    0      0     0     BMRU
eth1 1500    0     8268      0      0     0     14646   0      0     0     BMRU
kishore
  • 111
  • 1
  • 3
  • Sometimes a duplex mismatch on one end can look like this. Could you **paste into your question** the output of `netstat -in` on both machines? – MadHatter Mar 31 '14 at 06:16
  • Ifac MTUMet RX-ok rx-err rx-drp rx-ovr tx-ok tx-err tx-drp tx-ovr flg eth0 1500 0 3459 0 0 0 6175 0 0 0 for one machine and 1500 0 9906 0 0 0 5452 0 0 0 for the other machine – kishore Mar 31 '14 at 06:56
  • I've tried to fix up the formatting in your paste-in in the question, and either there's a column missing, or you have ONLY errors. I'm hoping that it's the former; could you check and adjust accordingly? – MadHatter Mar 31 '14 at 07:11
  • I performed netstat -in once again and I have updated the values which i obtained. I don't find any extra column. I assuned MTU Met to be a single field. IF that is not correct then I have to adjust accordingly. – kishore Mar 31 '14 at 07:16
  • @MadHatter I found out what i missed and edited the question accordingly. Is there anything wrong with those values? – kishore Mar 31 '14 at 07:28
  • 1
    Nope. The zero TX-err and TX-drp columns rule out duplex mismatch, pretty much. Oh well, it was worth checking. – MadHatter Mar 31 '14 at 07:56
  • Could the prob be with network card settings like large send offload or something like that? I have been trying too much without any success. Where could i possibly look for the cause? – kishore Mar 31 '14 at 08:07

0 Answers0