Stochastic Forecasts Achieve High Throughput and Low Delay over Cellular Networks

Transcription

1 Stochastic Forecasts Achieve High Throughput and Low Delay over Cellular Networks Keith Winstein MIT Computer Science and Artificial Intelligence Laboratory March 6, 2014 Joint work with Anirudh Sivaraman and Hari Balakrishnan.

2 Sprout: a transport protocol designed for variability Observation: Videoconferences perform poorly over cellular networks.

3 Verizon LTE uplink throughput 2000 throughput (kbps) :00 0:30 1:00 1:30

4 Verizon LTE ping delay during one TCP download 10 Round-trip time (s) Time (s)

5 Interactive apps work poorly We measured cellular networks while driving: Verizon LTE Verizon 3G (1xEV-DO) AT&T LTE T-Mobile 3G (UMTS) Then ran apps across replayed network trace: Skype (Windows 7) Google Hangouts (Chrome on Windows 7) Apple Facetime (OS X)

6 Skype s performance

7 Performance summary 8000 Verizon LTE Downlink 7000 Throughput (kbps) Facetime 2000 Better Skype Google Hangout Self-inflicted delay (ms)

8 What s wrong? Existing schemes react to congestion signals. Packet loss. Increase in round-trip time. Feedback comes too late. The killer: self-inflicted queueing delay.

9 Sprout s mission Most throughput Bounded risk of delay > 100 ms

10 Bounded risk of delay Model variation in link speed Infer current link speed Predict future link speed Don t wait for congestion Control: Send as much as possible, but require: 95% chance all packets arrive within 100 ms

11 Model: packet deliveries looks like flicker noise 100 (< 0.5) 10 t 3.27 Percent interarrivals interarrival time (ms) (Verizon LTE, phone stationary.)

12 Model: average rate looks like random walk 2000 throughput (kbps) :00 0:30 1:00 1:30

13 Sprout s model Poisson process drains queue Sender Queue Receiver Rate λ controls Poisson process λ Brownian motion of σ t varies λ σ If in an outage, λ z is escape rate. λ z

14 Sprout s model parameters Volatility σ: Expected outage time 1/λ z : 1 s Tick length (τ): 20 ms Forecast length: 160 ms Delay target: 100 ms Risk tolerance: 5% 200 pkts/s s All source code was frozen before data collection began.

15 Infer: current link speed Observe packets received every τ Update P(λ)

16 Predict: future link speed Evolve model forward Forecast 5th percentile cumulative packets

17 Control: fill up 100 ms forecast window Cumulative packets sent Cautious forecast time (ms)

18 Control: fill up 100 ms forecast window Cumulative packets sent ms Cautious forecast time (ms)

19 Control: fill up 100 ms forecast window Cumulative packets sent Send 50 pkts 100 ms Cautious forecast time (ms)

20 Control: fill up 100 ms forecast window 70 Cumulative packets sent Send 50 pkts 100 ms 100 ms Cautious forecast time (ms)

21 Control: fill up 100 ms forecast window 70 Cumulative packets sent Send 50 pkts Send ms 100 ms Cautious forecast time (ms)

22 Control: fill up 100 ms forecast window 70 Cumulative packets sent Send 50 pkts Send ms 100 ms 100 ms Cautious forecast time (ms)

23 Control: fill up 100 ms forecast window 70 Cumulative packets sent Send 50 pkts Send 5 Send ms 100 ms 100 ms Cautious forecast time (ms)

24 Sprout s results

25 Verizon LTE Downlink Throughput (kbps) Facetime 2000 Better Skype Google Hangout Self-inflicted delay (ms)

26 Verizon LTE Downlink Throughput (kbps) Better LEDBAT Vegas Compound TCP Facetime 2000 Skype Google Hangout Self-inflicted delay (ms)

27 Verizon LTE Downlink Throughput (kbps) Cubic Better LEDBAT Vegas Compound TCP Facetime 2000 Skype Google Hangout Self-inflicted delay (ms)

28 Verizon LTE Downlink Throughput (kbps) Cubic Better LEDBAT Sprout Vegas Compound TCP Facetime 2000 Skype Google Hangout Self-inflicted delay (ms)

29 Verizon LTE Downlink Sprout-EWMA Throughput (kbps) Cubic Better LEDBAT Sprout Vegas Compound TCP Facetime 2000 Skype Google Hangout Self-inflicted delay (ms)

30 Verizon LTE Uplink 5000 Sprout-EWMA Throughput (kbps) Cubic Skype Facetime Sprout LEDBAT Vegas Compound TCP Google Hangout Self-inflicted delay (ms)

31 Overall results on 8 links Verizon 3G/LTE, AT&T LTE, T-Mobile 3G uplink and downlink: Sprout vs. Avg. speedup Delay reduction Skype Hangout Facetime Compound TCP Vegas LEDBAT Same 2.8 Cubic

32 Sprout is end-to-end, but comparable to in-net control Sprout-EWMA Utilization (percent) Cubic Cubic-over-CoDel Sprout Self-inflicted delay (ms)

33 M.I.T contest (March April 2013) Turnkey network emulator, evaluation Sender, receiver run in Linux containers Mission: maximize throughput/delay 4th prize: $20 3rd prize: $30 2nd prize: $40 (If beat Sprout) 1st prize:

34 M.I.T contest (March April 2013) Turnkey network emulator, evaluation Sender, receiver run in Linux containers Mission: maximize throughput/delay 4th prize: $20 3rd prize: $30 2nd prize: $40 (If beat Sprout) 1st prize: Co-authorship on future paper

35 M.I.T contest (March April 2013) Turnkey network emulator, evaluation Sender, receiver run in Linux containers Mission: maximize throughput/delay 4th prize: $20 3rd prize: $30 2nd prize: $40 (If beat Sprout) 1st prize: Co-authorship on future paper Anirudh Sivaraman, KW, Pauline Varley, Somak Das, Joshua Ma, Ameesh Goyal, João Batalha, and Hari Balakrishnan, Protocol Design Contests, in submission

36 Baseline

37 Land of 3,000 student protocols

38 Sprout was on the frontier

39 Limitations Sprout wants to control all of the traffic on a queue. Cells generally have per-user queues but Wi-Fi and wired networks usually don t. What if cell link isn t the bottleneck? Assumption: application always has data to send

40 Our approach Pick a model, any model. All models are wrong, but they help anyway! See if it lands on the frontier.* * (On a large set of real network paths or newly-collected traces.) Kaizen for congestion

41 Thank you Lakshminarayanan Subramanian Shuo Deng Jonathan Perry Katrina LaCurts Andrew McGregor Tim Shepard Dave Täht Michael Welzl Hannes Tschofenig members ( NSF & Shannon family (fellowship)

42 Sprout for controlled delay over cellular networks Model varying link speed Infer current link speed Predict future link speed Control risk of large delay with cautious forecast Yields 2 4 throughput of Skype, Facetime, Hangout Achieves 7 9 reduction in self-inflicted delay Matches some active queue management purely end-to-end Code and directions at alfalfa@mit.edu

43 Verizon 3G (1xEV-DO) Downlink Cubic Throughput (kbps) Google Hangout Compound TCP Facetime Skype Sprout-EWMA LEDBAT Vegas Sprout Self-inflicted delay (ms)

44 Verizon 3G (1xEV-DO) Uplink 600 Sprout-EWMA Throughput (kbps) Cubic LEDBAT Skype Facetime Vegas Compound TCP Google Hangout Sprout Self-inflicted delay (ms)

45 AT&T LTE Downlink Sprout-EWMA Throughput (kbps) Cubic Sprout Facetime LEDBAT Compound TCP Vegas Skype Google Hangout Self-inflicted delay (ms)

46 AT&T LTE Uplink Throughput (kbps) Cubic LEDBAT Sprout-EWMA Compound TCP Vegas Skype Facetime Sprout 300 Google Hangout Self-inflicted delay (ms)

47 T-Mobile 3G (UMTS) Downlink Sprout-EWMA Throughput (kbps) Cubic LEDBAT Compound TCP Facetime Skype Google Hangout Vegas Sprout Self-inflicted delay (ms)

48 T-Mobile 3G (UMTS) Uplink Throughput (kbps) 1000 Cubic LEDBAT Sprout-EWMA Vegas Compound TCP Facetime Sprout Skype Google Hangout Self-inflicted delay (ms)