Sprout: Stochastic Forecasts Improve Performance in...

Sprout: Stochastic Forecasts ImprovePerformance in Cellular Networks

Keith Winstein, Anirudh Sivaraman, Hari Balakrishnan

6.829 Fall 2018

October 25, 2018

Cellular Network Architecture

http:

//image.slidesharecdn.com/introductiontomobilecorenetworkpublic-130325020435-phpapp01/95/

introduction-to-mobile-core-network-13-638.jpg

Keith Winstein, Anirudh Sivaraman, Hari Balakrishnan 6.829 Fall 2018

Sprout: Stochastic Forecasts Improve Performance in Cellular Networks

http://image.slidesharecdn.com/introductiontomobilecorenetworkpublic-130325020435-phpapp01/95/introduction-to-mobile-core-network-13-638.jpg



Wireless: Highly Variable Rates

0

500

1000

1500

2000

0:00 0:30 1:00 1:30

thro

ugh

pu

t(k

bp

s)

Verizon LTE uplink throughput



Awful Delays: The Too-Reliable Network

6.02 Fall 2012 Lecture 21, Slide #16 http://nms.csail.mit.edu/papers/index.php?detail=208

Verizon LTE TCP RTT (Cambridge, MA)

6.02 Fall 2012 Lecture 21, Slide #17 17

Round-trip delay

Del

ay

(millise

con

ds)

AT&T Wireless on iPhone 3G

mu: 1697.2 ms stddev: 2346.5 ms min:155.6 ms max:12126.6 ms

Time (s)

RTT (m

s)

On the Acela Amtrak train (BOS à NY)



Highly Variable Signal-to-Noise Ratio (SNR)

0

10

20

30

0 1 2 3 4 5 6 7 8

Mea

sure

d S

NR

(d

B)

Time (s)

0 10 20 30

0 50 100 150 200 250 Time (ms)

Shaded area detail

10-7

10-5

10-3

10-1

0 1 2 3 4 5 6 7 8

Bit

err

or

rate

Time (s)

Slow walk (indoors, 2.4 GHz)

M. Vutukuru, K. Jamieson, HB, “Cross-Layer Wireless Bit Rate Adaptation”, SIGCOMM 2009.



Packet Scheduling

I What is the “fair” way to schedule wireless users?

I Case 1: User i has a rate ri packets/s.

I Case 2: ri (t) is a function of time.



Packet Scheduling

I Cellular networks maintain per-device queues because it allowsthe base station to trade-off between efficiency and fairness.

I Scheduling depends on the state of the channel to a user.



Proportional Fair Wireless Scheduler

I Let ri (t) be the current (“instantaneous”) rate and let Ri (t)be the value of time t of an EWMA-filtered average:

Ri (t + 1) = (1− α)Ri (t) + αri (t) if i = j ,

andRi (t + 1) = (1− α)Ri (t) otherwise

I Select j that maximizes ri (t)Ri (t)

.



Video & Conferencing over Wireless

I We measured cellular networks while driving:

I Verizon LTEI Verizon 3G (1xEV-DO)I AT&T LTEI T-Mobile 3G (UMTS)

I Then ran apps across emulated network:

I Skype (Windows 7)I Google Hangout (Chrome on Windows 7)I Apple Facetime (OS X)



Characterizing Performance

0

500

1000

1500

2000

2500

3000

3500

300500100020003000500010000

Th

rou

gh

pu

t(k

bp

s)

Self-inflicted delay (ms)

Skype

Better



Why is Wireless Videoconferencing So Bad?

I Today’s protocols react to congestion signalsI Packet lossI Increase in round-trip time

I Feedback comes too late to help

I The killer: self-inflicted queueing delay

I Any overshoot means a queue filling up with packets



Sprout’s Goal

I Maximize throughput, but

I Bounded risk of delay > D (e.g., D = 100 ms).



Bounded Risk of Delay

I Infer rate from interarrival distribution

I Predict future link rate and convey prediction to sender

I Don’t wait for congestion

I Control: Send as fast as possible, but require:

I 95% probability all packets will arrive within 100 ms



Infer Rate from Interarrival Process

0.0001

0.001

0.01

0.1

1

10

100

(< 0.5)

39041 10 100 1000

Per

cen

tin

tera

rriv

als

interarrival time (ms)

t−3.27

Verizon LTE. Stationary phone. 3 am... “Flicker noise” process.Keith Winstein, Anirudh Sivaraman, Hari Balakrishnan 6.829 Fall 2018


Predict Future Link Rate

I Model rate evolution as random walk (Brownian motion)

I Count packets in every 20 ms tick

I Use Bayesian updating to make cautious forecast(5th percentile cumulative packets)

I Receiver makes forecast; tells sender in ACK



Network Model

Sender Queue Receiver

Poisson process

Rate λ controls

λ

σ

Brownian motion

λz

If in an outage,

drains queue

Poisson process

of σ√t varies λ λz is escape rate.



Bayesian Update

I Discrete set of possible rates, λ (e.g., 0 to 1000 packets/s)

I Initially, each λ is equi-probable

I Each tick (τ seconds), if we receive k bytes, run update step:

Pnew(λ = x)←Pold(λ = x) · (xτ)

k

k! e−(xτ)

Z,

where Z ensures that the probabilities sum to 1

I Pre-compute most of the math



The Cautious Forecast

I Receiver has “cloud” of current link speeds

I For eight ticks in the future:

I Predict future link rate by simulating Brownian motion of ratesI Find 5th percentile of cumulative packets

I Send forecast to sender on ACKs

I Most of the math is pre-computed



0

1000

2000

3000

4000

5000

6000

7000

8000

100200300500100020005000

Th

rou

gh

pu

t(k

bp

s)


Verizon LTE Downlink

SkypeFacetime

Google Hangout

Bette

r



0

1000

2000

3000

4000

5000

6000

7000

8000

100200300500100020005000

Th

rou

gh

pu

t(k

bp

s)



Compound TCP

LEDBAT

Cubic

SkypeFacetime

Google Hangout

Vegas

Bette

r



0

1000

2000

3000

4000

5000

6000

7000

8000

100200300500100020005000

Th

rou

gh

pu

t(k

bp

s)



Compound TCP

LEDBAT

Cubic

SkypeFacetime

Google Hangout

Vegas

Sprout

Bette

r



0

1000

2000

3000

4000

5000

6000

7000

8000

100200300500100020005000

Th

rou

gh

pu

t(k

bp

s)



Sprout-EWMA

Compound TCP

LEDBAT

Cubic

SkypeFacetime

Google Hangout

Vegas

Sprout

Bette

r



0

1000

2000

3000

4000

5000

300500100020003000500010000

Th

rou

gh

pu

t(k

bp

s)


Verizon LTE Uplink

Sprout-EWMA

Compound TCP

LEDBAT

Cubic

Skype Facetime

Google Hangout

Vegas

Sprout



150

200

250

300

350

400

450

500

550

200030005000100002000050000100000

Th

rou

gh

pu

t(k

bp

s)


Verizon 3G (1xEV-DO) Downlink

Sprout-EWMACompound TCP

LEDBAT

Cubic

Skype

FacetimeGoogle Hangout

Vegas

Sprout



200

300

400

500

600

200300500100020003000500010000

Th

rou

gh

pu

t(k

bp

s)


Verizon 3G (1xEV-DO) Uplink

Sprout-EWMA

Compound TCP

LEDBAT

Cubic

Skype

Facetime

Google Hangout

Vegas

Sprout



500

1000

1500

2000

2500

3000

3500

4000

3050100200300500100020005000

Th

rou

gh

pu

t(k

bp

s)


AT&T LTE Downlink

Sprout-EWMA

Compound TCPLEDBATCubic

Skype

Facetime

Google Hangout

Vegas

Sprout



200

300

400

500

600

700

800

900

2003005001000200050001000020000

Th

rou

gh

pu

t(k

bp

s)


AT&T LTE Uplink

Sprout-EWMA

Compound TCP

LEDBATCubic

SkypeFacetime

Google Hangout

Vegas

Sprout



200

400

600

800

1000

1200

1400

1600

3005001000200030005000

Th

rou

gh

pu

t(k

bp

s)


T-Mobile 3G (UMTS) Downlink

Sprout-EWMA

Compound TCPLEDBAT

Cubic

SkypeFacetime

Google Hangout

Vegas

Sprout



200

300

400

500

600

700

800

900

1000

2003005001000200050001000030000

Th

rou

gh

pu

t(k

bp

s)


T-Mobile 3G (UMTS) Uplink

Sprout-EWMA

Compound TCP

LEDBAT

Cubic

Skype

Facetime

Google Hangout

Vegas

Sprout



Overall results

Sprout vs. Avg. speedup Delay reduction

Skype 2.2× 7.9×Hangout 4.4× 7.2×Facetime 1.9× 8.7×Compound 1.3× 4.8×TCP Vegas 1.1× 2.1×LEDBAT Same 2.8×Cubic 0.91× 79×



Varying risk tolerance

200

300

400

500

600

700

800

900

1000

2003005001000300050001000030000

Th

rou

gh

pu

t(k

bp

s)


Compound TCP

LEDBAT

Cubic

Skype

Facetime

Google Hangout

Vegas

Sprout (5%)

25%

50%

75%

95%

T-Mobile 3G (UMTS) Uplink



Competes with Active Queue Mgmt (AQM)even though end-to-end

0

20

40

60

80

100

2003005001000300050001000030000

Uti

lizat

ion

(per

cen

t)


Sprout

Sprout-EWMA

Cubic

Cubic-over-CoDel



Replication by Stanford students(February–March 2013)

I Alterman & Quach reproduced a few of our measurements

I http://ReproducingNetworkResearch.wordpress.com/2013/03/12/1216/

I Won best project award in Stanford networking class!



M.I.T. 6.829 contest (March–April 2013)

I Turnkey network emulator, evaluation

I Sender, receiver run in Linux containers

I Leaderboards



Baseline



Land of 3,000 student protocols



Sprout is on the frontier



Limitations

I Stochastic model has not been tuned

I Only evaluated long-running flows.

I All testing data from Boston.

I User should wrap competing flows inside Sprout.

I Designed for cellular link with per-user queues

I Fortunately, cells have per-device queues. . .I . . . but Wi-Fi generally doesn’t.

I What about when the cell link isn’t the bottleneck?



Thoughts on Methods

I Pick a model, any model.

I All models are (at some level) wrong, but they help anyway!

I See if it lands on the throughput-delay frontier.*

* (On a large set of real network paths or newly-collectedtraces.)



Conclusion: High Throughput + Controlled DelayAchievable over Variable Wireless Networks

I Infer link speed from interarrival distribution

I Predict future link speed

I Control risk of large delay with cautious forecast

I Yields 2–4× throughput of Skype, Facetime, Hangout

I Achieves 7–9× reduction in self-inflicted delay

I Matches active queue management without router changes

I Code and directions at http://alfalfa.mit.edu



Date post:	31-Dec-2019
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Sprout: Stochastic Forecasts Improve Performance in...

Documents