Hi Golam, can you please check whether the branch you used from the test tree was 'master' or 'dccp' (via 'git branch')? I have checked the figures below against a test here - I get similar figures (in the order of several 100 Kbps) when using the 'master' branch, but on the 'dccp' branch the results are (iperf 20sec runs): * CCID-2: 399 Mbps * CCID-3: 511 Mbps * TCP/cubic: 726 Mbps CCID-2 is indeed lagging behind, it is not perfect yet, but it is a factor of 1000 faster than the top 182 kbps experienced when using the 'master' branch. If you are experiencing these results on the 'dccp' branch of the test tree, the next step should be to check whether any loss occurs. You can look at the 'ssthresh' values and the way cwnd grows when using dccp_probe Another thing that may help would be to enable the debugging output. This needs to be enabled at compile time and at runtime: $ echo 1 > /sys/module/dccp/parameters/ccid2_debug Gerrit Quoting Golam Sarwar: | I have upgraded my test machines to 2.6.31 (released on 9-11-2009) | experimental DCCP tree. With CCID2, My problem is - I can not send enough | data even though both my test machines are connected to 10Mbit/s LAN. While | with CCID3, for the same 2 machines, with exact same configuration, I can | transfer data properly without being limited by anything. | | The following is my iperf output for CCID2: | =============================================== | [ 4] 0.0- 1.0 sec 31.9 KBytes 261 Kbits/sec | [ 4] 1.0- 2.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 2.0- 3.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 3.0- 4.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 4.0- 5.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 5.0- 6.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 6.0- 7.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 7.0- 8.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 8.0- 9.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 9.0-10.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 10.0-11.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 11.0-12.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 12.0-13.0 sec 2.77 KBytes 22.7 Kbits/sec | [ 4] 0.0-13.0 sec 65.2 KBytes 41.0 Kbits/sec | =============================================== | | The following is my iperf output for CCID3: | =============================================== | [ 4] local 203.143.170.102 port 5001 connected with 203.143.170.200 port 56657 | [ 4] 0.0- 1.0 sec 11549 KBytes 94606 Kbits/sec | [ 4] 1.0- 2.0 sec 11586 KBytes 94913 Kbits/sec | [ 4] 2.0- 3.0 sec 11568 KBytes 94765 Kbits/sec | [ 4] 3.0- 4.0 sec 11583 KBytes 94890 Kbits/sec | [ 4] 4.0- 5.0 sec 11585 KBytes 94901 Kbits/sec | [ 4] 5.0- 6.0 sec 11576 KBytes 94833 Kbits/sec | [ 4] 6.0- 7.0 sec 11579 KBytes 94856 Kbits/sec | [ 4] 7.0- 8.0 sec 11583 KBytes 94890 Kbits/sec | [ 4] 8.0- 9.0 sec 11586 KBytes 94913 Kbits/sec | [ 4] 9.0-10.0 sec 11564 KBytes 94731 Kbits/sec | [ 4] 0.0-10.0 sec 115801 KBytes 94810 Kbits/sec | =============================================== | | I am using the following /proc entries to switch between CCID2 and CCID3: | =============================================== | sysctl -w net.dccp.default.seq_window=10000 | sysctl -w net.dccp.default.rx_ccid=2 | sysctl -w net.dccp.default.tx_ccid=2 | =============================================== | | I have double checked this issue with an older kernel (2.6.25-rc8) on the | same two machines and I don't get this problem with CCID2. The following is | the iperf receiving end output from one of the two 2.6.25-rc8 machines: | | =============================================== | [ 4] 0.0- 1.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 1.0- 2.0 sec 11508 KBytes 94277 Kbits/sec | [ 4] 2.0- 3.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 3.0- 4.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 4.0- 5.0 sec 11508 KBytes 94277 Kbits/sec | [ 4] 5.0- 6.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 6.0- 7.0 sec 11508 KBytes 94277 Kbits/sec | [ 4] 7.0- 8.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 8.0- 9.0 sec 11507 KBytes 94265 Kbits/sec | [ 4] 9.0-10.0 sec 11508 KBytes 94277 Kbits/sec | [ 4] 0.0-10.0 sec 115132 KBytes 94269 Kbits/sec | =============================================== -- To unsubscribe from this list: send the line "unsubscribe dccp" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
