While bisecting I found that the DCCP packet scheduler now reacts much differently from the one 2.6.20 (which was up to three times too fast). The new experimental data, taken on a 2.6.22 tree with the DCCP test tree patches, shows that (1) the packet scheduler now reacts accurately with regard to the value of t_ipi; (2) the send-credits patch is no longer necessary and in fact degrades performance. I would therefore like to drop this patch from the test tree. The results are here: http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/packet_scheduling/2.6.22-scheduler/ The most drastic example was a laptop with HZ=250 * 2.6.22 netdev as of 20/7: 84.6 Mbits/sec * DCCP test tree as of 20/7: 4.33 Mbits/sec !!! * DCCP test tree patched: 92.8 Mbits/sec I would be grateful for any test results and continue to look for further causes of the bad performance with the TBF bandwidth restriction. Any input, suggestions, test reports are welcome. - 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
