Kathie Nichols CoDel. present by Van Jacobson to the IETF-84 Transport Area Open Meeting 30 July 2012 Vancouver, Canada

Transcription

1 Kathie Nichols CoDel present by Van Jacobson to the IETF-84 Transport Area Open Meeting 30 July 2012 Vancouver, Canada

2 2

3 3

4 Sender Receiver 4

5 Sender Receiver 5

6 Sender Receiver Queue forms at a bottleneck There s probably just one bottleneck (each flow sees exactly one) Choices: can move the queue (by making a new bottleneck) or control it. 5

7 Good Queue / Bad Queue Queue length Time 6

8 Good Queue / Bad Queue Queue length Time Queue length Time 6

9 Good Queue / Bad Queue Queue length Good queue goes away in an RTT, bad queue hangs around. Time queue length min() over a sliding window measures bad queue... Queue length... as long as window is at least an RTT wide. Time 7

10 Good Queue / Bad Queue Queue length Good queue goes away in an RTT, bad queue hangs around. Time tracking min() in a sliding window gives bad queue... Queue length... as long as window is at least an RTT wide. Time 8

11 Good Queue / Bad Queue Queue length Good queue goes away in an RTT, bad queue hangs around. Time tracking min() in a sliding window gives bad queue... Queue length... as long as window is at least an RTT wide. Time 8

12 How big is the queue? Can measure size in bytes interesting if worried about overflow requires output bandwidth to compute delay Can measure packet s sojourn time direct measure of delay easy (no enque/deque coupling so works with any packet pipeline). 9

13 Sojourn Time Works with time-varying output bandwidth (e.g., wireless and shared links) Better behaved than queue length no high frequency phase noise Includes everything that affects packet so works for multi-queue links 10

14 Two views of a Queue Q delay (ms.) Top graph is sojourn time, bottom is queue size. Time (sec.) (one ftp + web traffic; 10Mbps bottleneck; 80ms RTT; TCP Reno) Q size (ms.)

15 Two views of a Queue Q delay (ms.) Top graph is sojourn time, bottom is queue size. Time (sec.) (one ftp + web traffic; 10Mbps bottleneck; 80ms RTT; TCP Reno) Q size (ms.)

16 Two views of a Queue Q delay (ms.) Top graph is sojourn time, bottom is queue size. (one ftp + web traffic; 10Mbps bottleneck; 80ms RTT; TCP Reno) Q size (ms.) Time (sec.) 12

17 Multi-Queue behavior 13

18 Controlling Queue a) Measure what you ve got b) Decide what you want c) If (a) isn t (b), move it toward (b) 14

19 Controlling Queue a) Measure what you ve got b) Decide what you want c) If (a) isn t (b), move it toward (b) - Estimator - Setpoint - Control loop 15

20 How much bad queue do we want? Can t let the link go idle (need one or two MTU of backlog) More than this will give higher utilization at low degree of multiplexing (1-3 bulk xfers) at the cost of higher delay Can the trade-off be quantified? 16

21 Utilization vs. Target for a single Reno TCP Bottleneck Link Utilization (% of max) Target (% of RTT) 17

22 Power vs. Target for a Reno TCP Average Power (Xput/Delay) Target (as % of RTT) 18

23 Power vs. Target for a Reno TCP Average Power (Xput/Delay) Target (as % of RTT) 18

24 Power vs. Target for a Reno TCP Average Power (Xput/Delay) Target (as % of RTT) 19

25 Power vs. Target for a Reno TCP Average Power (Xput/Delay) Target (as % of RTT) 19

26 20

27 Setpoint target of 5% of nominal RTT (5ms for 100ms RTT) yields substantial utilization improvement for small added delay. 20

28 Setpoint target of 5% of nominal RTT (5ms for 100ms RTT) yields substantial utilization improvement for small added delay. Result holds independent of bandwidth and congestion control algorithm (tested with Reno, Cubic & Westwood). 20

29 Setpoint target of 5% of nominal RTT (5ms for 100ms RTT) yields substantial utilization improvement for small added delay. Result holds independent of bandwidth and congestion control algorithm (tested with Reno, Cubic & Westwood). CoDel has no free parameters: runningmin window width determined by worstcase expected RTT and target is a fixed fraction of same RTT. 20

30 Algorithm & Control Law (see I-D, CACM paper and Linux kernels >= 3.5) 21

31 Eric Dumazet has combined CoDel with a simple SFQ ( buckets with RR service discipline). Cost in state & cycles is small and improvement is big. provides isolation - protects low rate CBR and web for a better user experience. Makes IW10 concerns a non-issue. gets rid of bottleneck bi-directional traffic problems ( ack-compression burstiness) improves flow mixing for better network performance (reduce HoL blocking) Since we re adding code, add fqcodel, not codel. 22

32 Where are we? thanks to Jim Gettys and the ACM, have dead-tree publication to protect ideas un-encumbered code (BSD/GPL dual-license) available for ns2, ns3 & linux in both simulation and real deployment, CoDel does no harm it either does nothing or reduces delay without affecting xput. 23

33 What needs to be done Still looking at parts of the algorithm but changes likely to be 2nd order. Would like to see CoDel on both ends of every home/small-office access link but: - We need to know more about how traffic behaves on particular bottlenecks (wi-fi, 3G cellular, cable modem) - There are system issues with deployment 24

34 Deployment Issues Home Gateway RTR/AP Cable Modem Headend 25

35 Deployment Issues Home Gateway RTR/AP Cable Modem Headend Linux kernel Protocol stack Device Driver Device 26

36 Deployment Issues Home Gateway RTR/AP Cable Modem Headend Linux kernel Protocol stack Device Driver Device Cellphone Phone CPU 3G Modem RAN? SGSN 27

37 Our thanks to: Jim Gettys CoDel early experimenters, particularly Dave Taht Eric Dumazet ACM Queue Eben Moglen 28