Hi Jonathan,
> On Mar 6, 2018, at 12:46, Jonathan Morton <[email protected]> wrote: > >> On 6 Mar, 2018, at 1:17 pm, Toke Høiland-Jørgensen <[email protected]> wrote: >> >>> So far it looks like everything is working fine and behaving as it >>> should (on a plain Ethernet interface). The one wrinkle is that it >>> takes a few thousand packets for the avg_off value to converge on the >>> true value, but that should easily be tolerable in most cases. >> >> So, erm, what do the new stats mean? ;) > > avg_off: the average observed offset of the transport header in packets. > This should converge to 14 on Ethernet. > > max/min_tran: tracking the transport-layer size of packets; max_tran = > max_len-avg_off if GSO is off. > > max/min_adj: tracking the overhead-adjusted size of packets; these should be > the adjusted sizes of the corresponding transport-layer sizes, so can be used > to check the calculations. > > It's probably feasible to lay these out better in the output, so they take up > fewer lines on screen. Great this matches what I deduced from the numbers. I note that by replacing a cake instance with itself (so not changing the configuration) will reset the counters. Question: am I right to assume that the _adj values are what you pass in as the true packet size into cake's calculation? I really like these outputs so far (running flent I saw: qdisc cake 8003: dev eth0 root refcnt 6 unlimited diffserv3 triple-isolate rtt 100.0ms noatm overhead 34 mpu 64 Sent 269164303 bytes 611190 pkt (dropped 0, overlimits 0 requeues 2) backlog 0b 0p requeues 2 memory used: 10880b of 15140Kb capacity estimate: 0bit Bulk Best Effort Voice thresh 0bit 0bit 0bit target 5.0ms 5.0ms 5.0ms interval 100.0ms 100.0ms 100.0ms pk_delay 14us 11us 8us av_delay 2us 2us 1us sp_delay 0us 0us 0us pkts 61400 395326 154464 bytes 17617151 183250552 68296600 way_inds 0 0 0 way_miss 3 89 11 way_cols 0 0 0 drops 0 0 0 marks 0 0 0 ack_drop 0 0 0 sp_flows 2 13 7 bk_flows 0 0 0 un_flows 0 0 0 max_len 3012 3028 3012 max_tran 1492 max_adj 1526 min_tran 28 min_adj 64 avg_off 14 Which is pretty much what I expected (speedtest ran over an effective MTU1492 PPPoE link*), but I still want to make more in-depth tests. Now we just need to carry this nto lede to get more testers. Plus it might make sense to merge tc-adv with iproute2-cake-next so that we have one master repository for the tc changes as well. I believe that before the xstats and Jonathan's recent updates iproute2-cake-next was more up to date, but I am probably wrong... Best Regards Sebastian *) Interestingly takling stats later revealed: qdisc cake 8003: dev eth0 root refcnt 6 unlimited diffserv3 triple-isolate rtt 100.0ms noatm overhead 34 mpu 64 Sent 743833445 bytes 1965042 pkt (dropped 0, overlimits 0 requeues 6) backlog 0b 0p requeues 6 memory used: 16346b of 15140Kb capacity estimate: 0bit Bulk Best Effort Voice thresh 0bit 0bit 0bit target 5.0ms 5.0ms 5.0ms interval 100.0ms 100.0ms 100.0ms pk_delay 8us 12us 9us av_delay 1us 3us 3us sp_delay 1us 1us 1us pkts 167869 1382188 414985 bytes 49022886 511662935 183147624 way_inds 0 341 0 way_miss 6 1137 20 way_cols 0 0 0 drops 0 0 0 marks 0 0 0 ack_drop 0 0 0 sp_flows 1 1 0 bk_flows 0 0 1 un_flows 0 0 0 max_len 3012 3028 3012 max_tran 1500 max_adj 1534 min_tran 28 min_adj 64 avg_off 14 Son between this sample and the speedtest the host exchanged MTU1500 packets with the local network... > > - Jonathan Morton > _______________________________________________ Cake mailing list [email protected] https://lists.bufferbloat.net/listinfo/cake
