TCP ex Machina: Computer-Generated Congestion Control Keith - - PowerPoint PPT Presentation

tcp ex machina computer generated congestion control
SMART_READER_LITE
LIVE PREVIEW

TCP ex Machina: Computer-Generated Congestion Control Keith - - PowerPoint PPT Presentation

TCP ex Machina: Computer-Generated Congestion Control Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory http://web.mit.edu/remy August 14, 2013 Introduction The problem Remy Evaluation


slide-1
SLIDE 1

TCP ex Machina: Computer-Generated Congestion Control

Keith Winstein and Hari Balakrishnan

MIT Computer Science and Artificial Intelligence Laboratory http://web.mit.edu/remy

August 14, 2013

slide-2
SLIDE 2

Introduction The problem Remy Evaluation Discussion

Congestion control!

◮ Prevents congestion collapse ◮ Allocates network resources among users ◮ Can be purely end-to-end or not

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-3
SLIDE 3

Introduction The problem Remy Evaluation Discussion

The march of congestion control mechanisms

DECbit Tahoe CARD DUAL

In-net End-to-end

1980s

GPS WFQ

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-4
SLIDE 4

Introduction The problem Remy Evaluation Discussion

The march of congestion control mechanisms

DECbit Tahoe CARD DUAL Reno NewReno RED BLUE Vegas ECN SACK

In-net End-to-end

1980s 1990s

GPS WFQ

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-5
SLIDE 5

Introduction The problem Remy Evaluation Discussion

The march of congestion control mechanisms

DECbit Tahoe CARD DUAL Reno NewReno RED BLUE Vegas ECN SACK EBCC CHOKe Westwood Binomial AVQ XCP BIC H-TCP Compound FAST VCP RCP Cubic

In-net End-to-end

Veno Eifel

1980s 1990s 2000s

GPS WFQ

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-6
SLIDE 6

Introduction The problem Remy Evaluation Discussion

The march of congestion control mechanisms

DECbit Tahoe CARD DUAL Reno NewReno RED BLUE Vegas ECN SACK EBCC CHOKe Westwood Binomial AVQ XCP BIC H-TCP Compound FAST VCP RCP Cubic DCTCP CoDel PDQ Sprout PRR

In-net End-to-end

D²TCP Veno Eifel

1980s 1990s 2000s 2010s

DeTail LEDBAT GPS WFQ

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-7
SLIDE 7

Introduction The problem Remy Evaluation Discussion

The march of congestion control mechanisms

DECbit Tahoe CARD DUAL Reno NewReno RED BLUE Vegas ECN SACK EBCC CHOKe Westwood Binomial AVQ XCP BIC H-TCP Compound FAST VCP RCP Cubic DCTCP CoDel PDQ Sprout FCP pFabric PRR GentleAggression

In-net End-to-end

D²TCP Veno Eifel

1980s 1990s 2000s 2010s

DeTail LEDBAT GPS WFQ

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-8
SLIDE 8

Introduction The problem Remy Evaluation Discussion

Our work

If congestion control is the answer, what’s the question?

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-9
SLIDE 9

Introduction The problem Remy Evaluation Discussion

Our work

If congestion control is the answer, what’s the question? Are there better answers?

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-10
SLIDE 10

Introduction The problem Remy Evaluation Discussion

Rational choice of scheme is challenging

Cubic

vs.

Compound

◮ Different goals? ◮ Different assumptions about network? ◮ One scheme just plain better?

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-11
SLIDE 11

Introduction The problem Remy Evaluation Discussion

Networks constrained by a fuzzy idea of TCP’s assumptions

◮ Mask stochastic loss ◮ Bufferbloat ◮ Mask out-of-order delivery ◮ No parallel/multipath routing

Advice for Internet Subnetwork Designers (RFC 3819) is 21,000 words!

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-12
SLIDE 12

Introduction The problem Remy Evaluation Discussion

Apps hack around TCP

◮ Open lots of flows ◮ Goose slow start ◮ Add pacing ◮ Give up and do it yourself

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-13
SLIDE 13

Introduction The problem Remy Evaluation Discussion

Apps hack around TCP

◮ Open lots of flows ◮ Goose slow start ◮ Add pacing ◮ Give up and do it yourself

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-14
SLIDE 14

Introduction The problem Remy Evaluation Discussion

Apps hack around TCP

◮ Open lots of flows ◮ Goose slow start ◮ Add pacing ◮ Give up and do it yourself

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-15
SLIDE 15

Introduction The problem Remy Evaluation Discussion

Apps hack around TCP

◮ Open lots of flows ◮ Goose slow start ◮ Add pacing ◮ Give up and do it yourself

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-16
SLIDE 16

Introduction The problem Remy Evaluation Discussion

Apps hack around TCP

◮ Open lots of flows ◮ Goose slow start ◮ Add pacing ◮ Give up and do it yourself

Chrome (QUIC) BitTorrent (µTP) Mosh (SSP) Aspera (fasp)

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-17
SLIDE 17

Introduction The problem Remy Evaluation Discussion

Better: free the network to evolve Transport layer should adapt to whatever:

◮ network does ◮ application wants

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-18
SLIDE 18

Introduction The problem Remy Evaluation Discussion

What we built Remy: a program that generates congestion-control schemes offline

Input:

◮ Prior assumptions

(what network may do)

◮ Goal

(what app wants) Output: CC algorithm for a TCP sender (RemyCC) Time: a few hours Cost: $5–$10 on Amazon EC2

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-19
SLIDE 19

Introduction The problem Remy Evaluation Discussion

The basic question of congestion control

At this moment, do I:

◮ send a packet ◮ not send a packet?

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-20
SLIDE 20

Introduction The problem Remy Evaluation Discussion

Objectives of congestion control

Maximize

  • i

log [throughputi] (proportionally fair throughput)

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-21
SLIDE 21

Introduction The problem Remy Evaluation Discussion

Objectives of congestion control

Maximize

  • i

log [throughputi] (proportionally fair throughput)

  • i

log

  • throughputi

delayi

  • (proportionally fair throughput/delay)

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-22
SLIDE 22

Introduction The problem Remy Evaluation Discussion

Objectives of congestion control

Maximize

  • i

log [throughputi] (proportionally fair throughput)

  • i

log

  • throughputi
  • delayi

δ

  • (proportionally fair throughput/delay)

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-23
SLIDE 23

Introduction The problem Remy Evaluation Discussion

Objectives of congestion control

Maximize

  • i

log [throughputi] (proportionally fair throughput)

  • i

log

  • throughputi
  • delayi

δ

  • (proportionally fair throughput/delay)

◮ mini throughputi

(max-min throughput) Minimize

◮ average flow completion time ◮ page load time ◮ tail completion time

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-24
SLIDE 24

Introduction The problem Remy Evaluation Discussion

Prior assumptions

◮ Model of network uncertainty

◮ Link speed distribution ◮ Delay distribution

◮ Traffic model

◮ Web browsing, MapReduce, videoconferencing Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-25
SLIDE 25

Introduction The problem Remy Evaluation Discussion

Dumbbell network

Sender Sender 2 Queue Receiver Round-trip time Link

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-26
SLIDE 26

Introduction The problem Remy Evaluation Discussion

Dumbbell network

Sender Sender 2 Queue Receiver Receiver 2 Round-trip time Link

? ?

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-27
SLIDE 27

Introduction The problem Remy Evaluation Discussion

Dumbbell network

...

Sender Sender 2 Sender n Queue

...

Receiver Receiver 2 Receiver n Round-trip time Link

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-28
SLIDE 28

Introduction The problem Remy Evaluation Discussion

Superrational congestion control

At this moment,* do I:

◮ send a packet ◮ not send a packet?

* Assuming every node is running the same algorithm.

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-29
SLIDE 29

Introduction The problem Remy Evaluation Discussion

Remy: search for superrationality

◮ Remy searches for the best congestion-control algorithm ◮ Optimizes expected objective over prior assumptions ◮ Makes tractable by limiting available state

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-30
SLIDE 30

Introduction The problem Remy Evaluation Discussion

A RemyCC tracks three congestion signals

r ewma: moving average of interval between acks s ewma: . . . between sender timestamps echoed in acks rtt ratio: ratio of last RTT to smallest RTT so far

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-31
SLIDE 31

Introduction The problem Remy Evaluation Discussion

Why these three congestion signals?

◮ Benefit can be measured empirically

◮ In our experiments, little help from adding more ◮ Other networks might find differently

◮ More signals increase search time

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-32
SLIDE 32

Introduction The problem Remy Evaluation Discussion

A RemyCC maps each state to an action Rule(r ewma, s ewma, rtt ratio) → m, b, τ m Multiple to congestion window b Increment to congestion window τ Minimum interval between two outgoing packets

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-33
SLIDE 33

Introduction The problem Remy Evaluation Discussion

Runtime for a RemyCC

On ack:

◮ m, b, τ ← Rule(r ewma, s ewma, rtt ratio) ◮ cwnd ← m · cwnd + b

Send packet if:

◮ cwnd > FlightSize, and ◮ last packet sent > τ ago

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-34
SLIDE 34

Introduction The problem Remy Evaluation Discussion

Remy’s job Find piecewise-continuous Rule() that optimizes expected value of objective function.

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-35
SLIDE 35

Introduction The problem Remy Evaluation Discussion

Remy example: 2D state space

On ack:

◮ m, b, τ ← Rule(r ewma, s ewma, rtt ratio )

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-36
SLIDE 36

Introduction The problem Remy Evaluation Discussion

Remy example: 2D state space

On ack:

◮ m, b, τ ← Rule(r ewma, s ewma, rtt ratio )

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-37
SLIDE 37

Introduction The problem Remy Evaluation Discussion

Remy example: Prior assumptions

Quantity Distribution Units Link speed Uniform(10, 20) Mbps RTT Uniform(100, 200) ms n Uniform(1, 16) “On” process exp[µ = 5] seconds “Off” process same

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-38
SLIDE 38

Introduction The problem Remy Evaluation Discussion

Remy example: Objective

  • i

log throughputi delayi

  • Keith Winstein and Hari Balakrishnan

MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-39
SLIDE 39

Introduction The problem Remy Evaluation Discussion

One action for all states. Find the best value.

s_ewma r_ewma

<?,?,?>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-40
SLIDE 40

Introduction The problem Remy Evaluation Discussion

The best (single) action. Now split it on median.

s_ewma r_ewma

<0.90,4,3.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-41
SLIDE 41

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.90,4,3.3> <0.90,4,3.3>

<0.90,4,3.3> <0.90,4,3.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-42
SLIDE 42

Introduction The problem Remy Evaluation Discussion

Optimize each of the new actions

s_ewma r_ewma

<0.90,4,3.3> <0.90,4,3.3>

<0.90,4,3.3> <0.90,4,3.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-43
SLIDE 43

Introduction The problem Remy Evaluation Discussion

Now split the most-used rule

s_ewma r_ewma

<0.90,5,2.8> <0.60,19,76.2>

<0.70,6,53.5> <0.80,5,4.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-44
SLIDE 44

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.90,5,2.8> <0.60,19,76.2>

<0.70,6,53.5>

<0.80,5,4.1> <0.80,5,4.1>

<0.80,5,4.1> <0.80,5,4.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-45
SLIDE 45

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.90,5,2.8> <0.60,19,76.2>

<0.70,6,53.5>

<0.80,5,4.1> <0.80,5,4.1>

<0.80,5,4.1> <0.80,5,4.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-46
SLIDE 46

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<0.90,5,2.8> <0.30,29,49.7>

<0.60,17,13.3>

<0.80,8,3.3> <0.80,8,62.7>

<0.80,17,4.6> <0.80,7,16.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-47
SLIDE 47

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.30,29,49.7>

<0.60,17,13.3>

<0.80,8,3.3> <0.80,8,62.7>

<0.80,17,4.6> <0.80,7,16.9>

<0.90,5,2.8> <0.90,5,2.8>

<0.90,5,2.8> <0.90,5,2.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-48
SLIDE 48

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.90,5,2.8> <0.90,5,2.8>

<0.90,5,2.8> <0.90,5,2.8>

<0.30,29,49.7>

<0.60,17,13.3>

<0.80,8,3.3> <0.80,8,62.7>

<0.80,17,4.6> <0.80,7,16.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-49
SLIDE 49

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.30,213,8.9> <0.90,-84,9.2>

<1.00,-139,9.1> <1.40,32,3.3>

<0.60,40,42.4>

<0.80,30,6.8>

<0.50,84,3.4> <0.60,8,57.2>

<0.90,-31,2.8> <1.30,184,15.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-50
SLIDE 50

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.30,213,8.9> <0.90,-84,9.2>

<1.00,-139,9.1>

<0.60,40,42.4>

<0.80,30,6.8>

<0.50,84,3.4> <0.60,8,57.2>

<0.90,-31,2.8> <1.30,184,15.3>

<1.40,32,3.3> <1.40,32,3.3> <1.40,32,3.3> <1.40,32,3.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-51
SLIDE 51

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.30,213,8.9> <0.90,-84,9.2>

<1.00,-139,9.1> <1.40,32,3.3> <1.40,32,3.3> <1.40,32,3.3> <1.40,32,3.3>

<0.60,40,42.4>

<0.80,30,6.8>

<0.50,84,3.4> <0.60,8,57.2>

<0.90,-31,2.8> <1.30,184,15.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-52
SLIDE 52

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.20,221,9.0> <0.90,-92,17.1>

<1.00,-220,2.8> <1.20,170,3.0> <1.40,185,4.9> <1.60,186,3.0> <1.40,170,3.2>

<0.50,256,27.2>

<1.00,206,6.6>

<0.50,216,3.5> <0.80,250,56.3>

<1.00,-96,4.2> <0.90,201,16.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-53
SLIDE 53

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.20,221,9.0> <0.90,-92,17.1>

<1.00,-220,2.8> <1.40,185,4.9> <1.60,186,3.0> <1.40,170,3.2>

<0.50,256,27.2>

<1.00,206,6.6>

<0.50,216,3.5> <0.80,250,56.3>

<1.00,-96,4.2> <0.90,201,16.2>

<1.20,170,3.0> <1.20,170,3.0> <1.20,170,3.0> <1.20,170,3.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-54
SLIDE 54

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.20,221,9.0> <0.90,-92,17.1>

<1.00,-220,2.8> <1.20,170,3.0> <1.20,170,3.0> <1.20,170,3.0> <1.20,170,3.0> <1.40,185,4.9> <1.60,186,3.0> <1.40,170,3.2>

<0.50,256,27.2>

<1.00,206,6.6>

<0.50,216,3.5> <0.80,250,56.3>

<1.00,-96,4.2> <0.90,201,16.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-55
SLIDE 55

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.90,253,8.1> <0.70,-157,70.0>

<1.00,-254,3.8> <1.30,251,2.8> <1.00,254,4.6> <1.10,250,2.9> <1.40,187,2.9> <1.50,194,4.1> <1.80,255,2.8> <1.50,251,3.1>

<0.60,256,28.0>

<1.20,232,6.6>

<0.60,234,3.3> <0.80,253,71.4>

<0.80,-105,4.3> <0.80,203,16.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-56
SLIDE 56

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.90,253,8.1> <0.70,-157,70.0>

<1.00,-254,3.8> <1.00,254,4.6> <1.10,250,2.9> <1.40,187,2.9> <1.50,194,4.1> <1.80,255,2.8> <1.50,251,3.1>

<0.60,256,28.0>

<1.20,232,6.6>

<0.60,234,3.3> <0.80,253,71.4>

<0.80,-105,4.3> <0.80,203,16.3>

<1.30,251,2.8> <1.30,251,2.8> <1.30,251,2.8> <1.30,251,2.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-57
SLIDE 57

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.90,253,8.1> <0.70,-157,70.0>

<1.00,-254,3.8> <1.30,251,2.8> <1.30,251,2.8> <1.30,251,2.8> <1.30,251,2.8> <1.00,254,4.6> <1.10,250,2.9> <1.40,187,2.9> <1.50,194,4.1> <1.80,255,2.8> <1.50,251,3.1>

<0.60,256,28.0>

<1.20,232,6.6>

<0.60,234,3.3> <0.80,253,71.4>

<0.80,-105,4.3> <0.80,203,16.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-58
SLIDE 58

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.70,256,8.1> <0.80,-222,69.3>

<1.10,-255,4.7> <1.00,251,3.3> <1.10,251,2.6> <1.30,253,3.6> <1.10,255,2.8> <0.80,255,4.7> <1.40,252,2.9> <1.10,253,2.9> <1.20,237,4.0> <2.00,256,2.7> <1.50,254,2.9>

<0.50,256,27.3>

<1.20,248,6.5>

<0.70,252,3.2> <1.20,255,71.1>

<0.80,-243,3.6> <0.90,227,15.5>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-59
SLIDE 59

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.70,256,8.1> <0.80,-222,69.3>

<1.10,-255,4.7> <1.00,251,3.3> <1.10,251,2.6> <1.30,253,3.6> <1.10,255,2.8> <0.80,255,4.7> <1.40,252,2.9> <1.10,253,2.9> <1.20,237,4.0> <2.00,256,2.7> <1.50,254,2.9>

<0.50,256,27.3>

<1.20,248,6.5>

<0.70,252,3.2> <1.20,255,71.1>

<0.80,-243,3.6>

<0.90,227,15.5>

<0.90,227,15.5>

<0.90,227,15.5>

<0.90,227,15.5>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-60
SLIDE 60

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.70,256,8.1> <0.80,-222,69.3>

<1.10,-255,4.7> <1.00,251,3.3> <1.10,251,2.6> <1.30,253,3.6> <1.10,255,2.8> <0.80,255,4.7> <1.40,252,2.9> <1.10,253,2.9> <1.20,237,4.0> <2.00,256,2.7> <1.50,254,2.9>

<0.50,256,27.3>

<1.20,248,6.5>

<0.70,252,3.2> <1.20,255,71.1>

<0.80,-243,3.6>

<0.90,227,15.5>

<0.90,227,15.5>

<0.90,227,15.5>

<0.90,227,15.5>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-61
SLIDE 61

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.40,256,8.1> <0.70,-222,75.8>

<1.00,-256,5.6> <0.80,251,3.0> <1.10,251,3.0> <1.00,253,2.7> <1.10,256,2.9> <0.80,255,4.0> <1.10,256,2.9> <1.00,254,3.0> <1.00,253,3.8> <2.00,256,3.2> <1.60,256,3.1>

<0.60,256,23.5>

<0.80,256,6.6>

<0.50,253,3.7> <1.40,256,65.5>

<1.00,-253,2.8>

<1.20,244,13.8>

<0.80,227,37.3>

<1.00,227,66.8>

<1.00,227,13.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-62
SLIDE 62

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.40,256,8.1> <0.70,-222,75.8>

<1.00,-256,5.6> <0.80,251,3.0> <1.10,251,3.0> <1.00,253,2.7> <1.10,256,2.9> <0.80,255,4.0> <1.10,256,2.9> <1.00,253,3.8> <2.00,256,3.2> <1.60,256,3.1>

<0.60,256,23.5>

<0.80,256,6.6>

<0.50,253,3.7> <1.40,256,65.5>

<1.00,-253,2.8>

<1.20,244,13.8>

<0.80,227,37.3>

<1.00,227,66.8>

<1.00,227,13.8> <1.00,254,3.0> <1.00,254,3.0> <1.00,254,3.0> <1.00,254,3.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-63
SLIDE 63

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.40,256,8.1> <0.70,-222,75.8>

<1.00,-256,5.6> <0.80,251,3.0> <1.10,251,3.0> <1.00,253,2.7> <1.10,256,2.9> <0.80,255,4.0> <1.10,256,2.9> <1.00,254,3.0> <1.00,254,3.0> <1.00,254,3.0> <1.00,254,3.0> <1.00,253,3.8> <2.00,256,3.2> <1.60,256,3.1>

<0.60,256,23.5>

<0.80,256,6.6>

<0.50,253,3.7> <1.40,256,65.5>

<1.00,-253,2.8>

<1.20,244,13.8>

<0.80,227,37.3>

<1.00,227,66.8>

<1.00,227,13.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-64
SLIDE 64

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.80,256,8.7> <0.80,-230,19.1>

<1.00,-256,5.6> <0.80,251,4.0> <1.40,251,2.2> <1.30,253,3.7> <1.30,256,2.9> <1.10,256,4.5> <0.90,256,2.9> <1.40,254,3.0> <1.10,254,2.1> <1.10,254,4.7> <1.00,254,2.8> <1.10,253,3.8> <1.80,256,3.9> <1.40,256,2.9>

<0.80,256,23.4>

<0.50,256,5.8>

<0.70,254,3.7> <1.90,256,64.0>

<1.10,-256,3.6>

<1.10,254,13.8>

<0.90,227,45.2>

<0.80,227,7.7>

<1.10,227,25.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-65
SLIDE 65

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.80,256,8.7> <0.80,-230,19.1>

<1.00,-256,5.6> <0.80,251,4.0> <1.40,251,2.2> <1.30,253,3.7> <1.30,256,2.9> <1.10,256,4.5> <0.90,256,2.9> <1.40,254,3.0> <1.10,254,2.1> <1.10,254,4.7> <1.00,254,2.8> <1.10,253,3.8> <1.80,256,3.9> <1.40,256,2.9>

<0.80,256,23.4>

<0.50,256,5.8>

<1.90,256,64.0>

<1.10,-256,3.6>

<1.10,254,13.8>

<0.90,227,45.2>

<0.80,227,7.7>

<1.10,227,25.0> <0.70,254,3.7> <0.70,254,3.7>

<0.70,254,3.7> <0.70,254,3.7>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-66
SLIDE 66

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.80,256,8.7> <0.80,-230,19.1>

<1.00,-256,5.6> <0.80,251,4.0> <1.40,251,2.2> <1.30,253,3.7> <1.30,256,2.9> <1.10,256,4.5> <0.90,256,2.9> <1.40,254,3.0> <1.10,254,2.1> <1.10,254,4.7> <1.00,254,2.8> <1.10,253,3.8> <1.80,256,3.9> <1.40,256,2.9>

<0.80,256,23.4>

<0.50,256,5.8>

<0.70,254,3.7> <0.70,254,3.7>

<0.70,254,3.7> <0.70,254,3.7>

<1.90,256,64.0>

<1.10,-256,3.6>

<1.10,254,13.8>

<0.90,227,45.2>

<0.80,227,7.7>

<1.10,227,25.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-67
SLIDE 67

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.90,256,8.6> <0.80,-230,24.8>

<1.10,-256,6.7> <0.90,253,3.1> <1.50,251,3.1> <1.30,253,2.9> <0.90,256,2.9> <0.90,256,4.1> <1.10,256,2.1> <1.20,256,2.8> <1.30,255,3.6> <1.30,254,6.3> <1.00,256,2.9> <1.30,254,3.9> <1.80,256,3.0> <1.20,256,3.1>

<0.80,256,17.0>

<0.40,256,5.3>

<0.80,254,3.8> <0.50,256,16.4>

<1.00,255,3.8> <0.70,255,3.7>

<1.80,256,50.6>

<1.40,-256,4.2>

<0.90,254,13.1>

<0.90,227,52.4>

<0.80,227,7.1>

<1.00,227,32.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-68
SLIDE 68

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.90,256,8.6> <0.80,-230,24.8>

<1.10,-256,6.7> <0.90,253,3.1> <1.50,251,3.1> <1.30,253,2.9> <0.90,256,2.9> <0.90,256,4.1> <1.10,256,2.1> <1.20,256,2.8> <1.30,255,3.6> <1.30,254,6.3> <1.00,256,2.9> <1.30,254,3.9> <1.80,256,3.0>

<0.80,256,17.0>

<0.40,256,5.3>

<0.80,254,3.8> <0.50,256,16.4>

<1.00,255,3.8> <0.70,255,3.7>

<1.80,256,50.6>

<1.40,-256,4.2>

<0.90,254,13.1>

<0.90,227,52.4>

<0.80,227,7.1>

<1.00,227,32.0> <1.20,256,3.1> <1.20,256,3.1> <1.20,256,3.1> <1.20,256,3.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-69
SLIDE 69

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.90,256,8.6> <0.80,-230,24.8>

<1.10,-256,6.7> <0.90,253,3.1> <1.50,251,3.1> <1.30,253,2.9> <0.90,256,2.9> <0.90,256,4.1> <1.10,256,2.1> <1.20,256,2.8> <1.30,255,3.6> <1.30,254,6.3> <1.00,256,2.9> <1.30,254,3.9> <1.80,256,3.0> <1.20,256,3.1> <1.20,256,3.1> <1.20,256,3.1> <1.20,256,3.1>

<0.80,256,17.0>

<0.40,256,5.3>

<0.80,254,3.8> <0.50,256,16.4>

<1.00,255,3.8> <0.70,255,3.7>

<1.80,256,50.6>

<1.40,-256,4.2>

<0.90,254,13.1>

<0.90,227,52.4>

<0.80,227,7.1>

<1.00,227,32.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-70
SLIDE 70

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.30,256,8.5> <0.40,-247,11.2>

<1.00,-256,5.2> <0.90,253,3.0> <1.60,251,3.1> <1.30,253,3.1> <0.90,256,2.8> <0.70,256,4.0> <1.50,256,2.9> <1.30,256,2.9> <1.50,255,4.2> <1.20,254,3.9> <1.00,256,2.8> <1.50,254,4.1> <1.70,256,2.2> <1.30,256,3.1> <1.10,256,3.8> <1.10,256,4.9> <0.90,256,3.1>

<0.70,256,10.6>

<0.30,256,7.0>

<0.90,256,3.1> <0.60,256,28.5>

<0.80,256,9.1> <0.50,256,3.8>

<1.60,256,44.1>

<1.20,-256,3.2>

<0.70,256,13.4>

<0.90,227,57.9>

<1.00,227,14.3>

<1.00,227,24.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-71
SLIDE 71

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.30,256,8.5> <0.40,-247,11.2>

<1.00,-256,5.2> <0.90,253,3.0> <1.60,251,3.1> <1.30,253,3.1> <0.90,256,2.8> <0.70,256,4.0> <1.50,256,2.9> <1.30,256,2.9> <1.50,255,4.2> <1.20,254,3.9> <1.00,256,2.8> <1.50,254,4.1> <1.70,256,2.2> <1.30,256,3.1> <1.10,256,3.8> <1.10,256,4.9> <0.90,256,3.1>

<0.70,256,10.6>

<0.30,256,7.0>

<0.60,256,28.5>

<0.80,256,9.1> <0.50,256,3.8>

<1.60,256,44.1>

<1.20,-256,3.2>

<0.70,256,13.4>

<0.90,227,57.9>

<1.00,227,14.3>

<1.00,227,24.0> <0.90,256,3.1> <0.90,256,3.1> <0.90,256,3.1> <0.90,256,3.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-72
SLIDE 72

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.30,256,8.5> <0.40,-247,11.2>

<1.00,-256,5.2> <0.90,253,3.0> <1.60,251,3.1> <1.30,253,3.1> <0.90,256,2.8> <0.70,256,4.0> <1.50,256,2.9> <1.30,256,2.9> <1.50,255,4.2> <1.20,254,3.9> <1.00,256,2.8> <1.50,254,4.1> <1.70,256,2.2> <1.30,256,3.1> <1.10,256,3.8> <1.10,256,4.9> <0.90,256,3.1>

<0.70,256,10.6>

<0.30,256,7.0>

<0.90,256,3.1> <0.90,256,3.1> <0.90,256,3.1> <0.90,256,3.1> <0.60,256,28.5>

<0.80,256,9.1> <0.50,256,3.8>

<1.60,256,44.1>

<1.20,-256,3.2>

<0.70,256,13.4>

<0.90,227,57.9>

<1.00,227,14.3>

<1.00,227,24.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-73
SLIDE 73

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.30,256,7.8> <0.20,-256,5.7>

<1.20,-256,5.2> <0.50,255,3.0> <1.50,251,9.0> <1.50,254,3.1> <1.10,256,2.8> <0.70,256,3.9> <1.50,256,3.2> <1.10,256,3.1> <1.90,255,3.2> <1.60,254,3.2> <0.50,256,2.9> <1.60,254,4.0> <1.60,256,2.8> <1.20,256,2.9> <0.90,256,4.2> <1.50,256,6.3> <0.70,256,2.7>

<0.70,256,10.5>

<0.10,256,3.7>

<0.80,256,3.9> <1.20,256,3.9> <0.60,256,3.1> <1.00,256,3.7> <0.30,256,32.3>

<0.90,256,8.3> <0.50,256,4.0>

<1.50,256,44.0>

<1.10,-256,3.1>

<0.70,256,12.4>

<0.80,227,50.9>

<1.10,227,14.9>

<1.00,227,22.4>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-74
SLIDE 74

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.30,256,7.8> <0.20,-256,5.7>

<1.20,-256,5.2> <0.50,255,3.0> <1.50,251,9.0> <1.50,254,3.1> <1.10,256,2.8> <0.70,256,3.9> <1.50,256,3.2> <1.10,256,3.1> <1.90,255,3.2> <1.60,254,3.2> <0.50,256,2.9> <1.60,254,4.0> <1.60,256,2.8> <1.20,256,2.9> <0.90,256,4.2> <1.50,256,6.3> <0.70,256,2.7>

<0.70,256,10.5>

<0.10,256,3.7>

<0.80,256,3.9> <1.20,256,3.9> <0.60,256,3.1> <1.00,256,3.7> <0.30,256,32.3>

<0.90,256,8.3> <0.50,256,4.0>

<1.50,256,44.0>

<1.10,-256,3.1>

<0.80,227,50.9>

<1.10,227,14.9>

<1.00,227,22.4>

<0.70,256,12.4> <0.70,256,12.4>

<0.70,256,12.4> <0.70,256,12.4>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-75
SLIDE 75

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.30,256,7.8> <0.20,-256,5.7>

<1.20,-256,5.2> <0.50,255,3.0> <1.50,251,9.0> <1.50,254,3.1> <1.10,256,2.8> <0.70,256,3.9> <1.50,256,3.2> <1.10,256,3.1> <1.90,255,3.2> <1.60,254,3.2> <0.50,256,2.9> <1.60,254,4.0> <1.60,256,2.8> <1.20,256,2.9> <0.90,256,4.2> <1.50,256,6.3> <0.70,256,2.7>

<0.70,256,10.5>

<0.10,256,3.7>

<0.80,256,3.9> <1.20,256,3.9> <0.60,256,3.1> <1.00,256,3.7> <0.30,256,32.3>

<0.90,256,8.3> <0.50,256,4.0>

<1.50,256,44.0>

<1.10,-256,3.1>

<0.70,256,12.4> <0.70,256,12.4>

<0.70,256,12.4> <0.70,256,12.4> <0.80,227,50.9>

<1.10,227,14.9>

<1.00,227,22.4>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-76
SLIDE 76

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<0.90,256,7.9> <0.30,-256,10.3>

<0.70,-256,11.7> <0.50,255,4.6> <1.40,252,7.4> <1.40,254,3.0> <1.10,256,3.6> <0.50,256,2.4> <1.30,256,3.1> <0.70,256,2.9> <1.80,256,3.2> <1.50,254,3.4> <0.60,256,2.9> <1.90,255,4.4> <1.80,256,3.4> <0.90,256,3.2> <0.80,256,4.0> <1.30,256,5.6> <1.00,256,2.9>

<0.50,256,10.0>

<0.30,256,4.4>

<1.00,256,3.2> <1.30,256,3.8> <0.40,256,2.5> <1.20,256,4.4> <0.70,256,26.5>

<0.60,256,9.2> <0.50,256,4.2>

<1.30,256,32.7>

<1.20,-256,3.7>

<0.40,256,10.8> <0.50,256,26.2>

<0.50,256,3.4> <0.90,256,19.6> <0.80,227,57.3>

<1.50,227,60.6>

<1.20,227,27.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-77
SLIDE 77

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.90,256,7.9> <0.30,-256,10.3>

<0.70,-256,11.7> <0.50,255,4.6> <1.40,252,7.4> <1.40,254,3.0> <1.10,256,3.6> <0.50,256,2.4> <1.30,256,3.1> <0.70,256,2.9> <1.80,256,3.2> <1.50,254,3.4> <0.60,256,2.9> <1.90,255,4.4> <1.80,256,3.4> <0.90,256,3.2> <0.80,256,4.0> <1.30,256,5.6> <1.00,256,2.9>

<0.50,256,10.0>

<0.30,256,4.4>

<1.00,256,3.2> <1.30,256,3.8> <0.40,256,2.5> <1.20,256,4.4> <0.70,256,26.5>

<0.60,256,9.2>

<1.30,256,32.7>

<1.20,-256,3.7>

<0.40,256,10.8> <0.50,256,26.2>

<0.50,256,3.4> <0.90,256,19.6> <0.80,227,57.3>

<1.50,227,60.6>

<1.20,227,27.2> <0.50,256,4.2> <0.50,256,4.2> <0.50,256,4.2> <0.50,256,4.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-78
SLIDE 78

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.90,256,7.9> <0.30,-256,10.3>

<0.70,-256,11.7> <0.50,255,4.6> <1.40,252,7.4> <1.40,254,3.0> <1.10,256,3.6> <0.50,256,2.4> <1.30,256,3.1> <0.70,256,2.9> <1.80,256,3.2> <1.50,254,3.4> <0.60,256,2.9> <1.90,255,4.4> <1.80,256,3.4> <0.90,256,3.2> <0.80,256,4.0> <1.30,256,5.6> <1.00,256,2.9>

<0.50,256,10.0>

<0.30,256,4.4>

<1.00,256,3.2> <1.30,256,3.8> <0.40,256,2.5> <1.20,256,4.4> <0.70,256,26.5>

<0.60,256,9.2>

<0.50,256,4.2> <0.50,256,4.2> <0.50,256,4.2> <0.50,256,4.2>

<1.30,256,32.7>

<1.20,-256,3.7>

<0.40,256,10.8> <0.50,256,26.2>

<0.50,256,3.4> <0.90,256,19.6> <0.80,227,57.3>

<1.50,227,60.6>

<1.20,227,27.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-79
SLIDE 79

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<0.70,256,8.1> <0.30,-256,15.8>

<0.70,-256,17.1> <0.50,255,11.8> <1.40,252,53.8> <1.40,254,10.1> <1.00,256,5.4> <0.10,256,2.3> <1.20,256,3.9> <0.20,256,3.1> <1.60,256,4.0> <1.40,254,3.0> <0.70,256,3.1> <1.70,256,5.4> <1.80,256,3.3> <0.90,256,3.0> <1.40,256,4.1> <0.70,256,3.8> <0.90,256,3.5>

<0.70,256,9.2>

<0.70,256,3.6>

<0.90,256,3.2> <1.30,256,5.2> <0.30,256,4.3> <1.40,256,4.4> <0.50,256,26.7>

<0.40,256,8.4>

<0.50,256,4.0> <0.50,256,5.2> <0.40,256,4.5> <0.50,256,4.0>

<0.90,256,33.4>

<1.30,-256,4.3>

<0.90,256,10.7> <0.60,256,33.4>

<0.90,256,4.2> <0.90,256,12.3> <0.80,227,63.7>

<1.30,228,65.6>

<1.10,227,39.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-80
SLIDE 80

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.70,256,8.1> <0.30,-256,15.8>

<0.70,-256,17.1> <0.50,255,11.8> <1.40,252,53.8> <1.40,254,10.1> <1.00,256,5.4> <0.10,256,2.3> <1.20,256,3.9> <0.20,256,3.1> <1.60,256,4.0> <1.40,254,3.0> <0.70,256,3.1> <1.70,256,5.4> <1.80,256,3.3> <0.90,256,3.0> <1.40,256,4.1> <0.70,256,3.8> <0.90,256,3.5>

<0.70,256,9.2>

<0.70,256,3.6>

<0.90,256,3.2> <1.30,256,5.2> <0.30,256,4.3> <1.40,256,4.4> <0.50,256,26.7>

<0.40,256,8.4>

<0.50,256,4.0> <0.50,256,5.2> <0.40,256,4.5> <0.50,256,4.0>

<0.90,256,33.4>

<1.30,-256,4.3>

<0.60,256,33.4>

<0.90,256,4.2> <0.90,256,12.3> <0.80,227,63.7>

<1.30,228,65.6>

<1.10,227,39.1>

<0.90,256,10.7> <0.90,256,10.7>

<0.90,256,10.7> <0.90,256,10.7>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-81
SLIDE 81

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.70,256,8.1> <0.30,-256,15.8>

<0.70,-256,17.1> <0.50,255,11.8> <1.40,252,53.8> <1.40,254,10.1> <1.00,256,5.4> <0.10,256,2.3> <1.20,256,3.9> <0.20,256,3.1> <1.60,256,4.0> <1.40,254,3.0> <0.70,256,3.1> <1.70,256,5.4> <1.80,256,3.3> <0.90,256,3.0> <1.40,256,4.1> <0.70,256,3.8> <0.90,256,3.5>

<0.70,256,9.2>

<0.70,256,3.6>

<0.90,256,3.2> <1.30,256,5.2> <0.30,256,4.3> <1.40,256,4.4> <0.50,256,26.7>

<0.40,256,8.4>

<0.50,256,4.0> <0.50,256,5.2> <0.40,256,4.5> <0.50,256,4.0>

<0.90,256,33.4>

<1.30,-256,4.3>

<0.90,256,10.7> <0.90,256,10.7>

<0.90,256,10.7> <0.90,256,10.7>

<0.60,256,33.4>

<0.90,256,4.2> <0.90,256,12.3> <0.80,227,63.7>

<1.30,228,65.6>

<1.10,227,39.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-82
SLIDE 82

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<0.80,256,9.7> <0.40,-256,60.0>

<0.90,-256,24.3> <0.50,255,18.2> <1.50,252,4.2> <1.40,254,22.9> <1.00,256,57.3> <0.20,256,1.4> <1.20,256,7.8> <0.30,256,3.1> <1.50,256,7.9> <1.60,254,3.0> <1.00,256,3.2> <1.90,256,3.9> <1.90,256,3.1> <0.80,256,3.1> <1.50,256,4.9> <0.50,256,3.2> <0.70,256,3.5>

<0.70,256,8.5>

<0.70,256,4.6>

<0.80,256,3.3> <1.20,256,5.0> <0.70,256,3.5> <1.40,256,4.4> <0.40,256,21.3>

<0.10,256,7.5>

<0.50,256,3.8> <0.50,256,11.6> <0.30,256,10.8> <0.50,256,4.0>

<1.10,256,39.5>

<1.40,-256,3.7>

<1.20,256,9.1> <0.90,256,11.8>

<1.20,256,16.2> <0.70,256,7.3>

<0.80,256,39.9>

<1.00,256,5.0> <1.00,256,17.9> <0.70,227,8.5>

<1.40,228,28.2>

<1.10,227,56.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-83
SLIDE 83

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.80,256,9.7> <0.40,-256,60.0>

<0.90,-256,24.3> <0.50,255,18.2> <1.50,252,4.2> <1.40,254,22.9> <1.00,256,57.3> <0.20,256,1.4> <1.20,256,7.8> <0.30,256,3.1> <1.50,256,7.9> <1.60,254,3.0> <1.90,256,3.9> <1.90,256,3.1> <0.80,256,3.1> <1.50,256,4.9> <0.50,256,3.2> <0.70,256,3.5>

<0.70,256,8.5>

<0.70,256,4.6>

<0.80,256,3.3> <1.20,256,5.0> <0.70,256,3.5> <1.40,256,4.4> <0.40,256,21.3>

<0.10,256,7.5>

<0.50,256,3.8> <0.50,256,11.6> <0.30,256,10.8> <0.50,256,4.0>

<1.10,256,39.5>

<1.40,-256,3.7>

<1.20,256,9.1> <0.90,256,11.8>

<1.20,256,16.2> <0.70,256,7.3>

<0.80,256,39.9>

<1.00,256,5.0> <1.00,256,17.9> <0.70,227,8.5>

<1.40,228,28.2>

<1.10,227,56.0> <1.00,256,3.2> <1.00,256,3.2> <1.00,256,3.2> <1.00,256,3.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-84
SLIDE 84

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.80,256,9.7> <0.40,-256,60.0>

<0.90,-256,24.3> <0.50,255,18.2> <1.50,252,4.2> <1.40,254,22.9> <1.00,256,57.3> <0.20,256,1.4> <1.20,256,7.8> <0.30,256,3.1> <1.50,256,7.9> <1.60,254,3.0> <1.00,256,3.2> <1.00,256,3.2> <1.00,256,3.2> <1.00,256,3.2> <1.90,256,3.9> <1.90,256,3.1> <0.80,256,3.1> <1.50,256,4.9> <0.50,256,3.2> <0.70,256,3.5>

<0.70,256,8.5>

<0.70,256,4.6>

<0.80,256,3.3> <1.20,256,5.0> <0.70,256,3.5> <1.40,256,4.4> <0.40,256,21.3>

<0.10,256,7.5>

<0.50,256,3.8> <0.50,256,11.6> <0.30,256,10.8> <0.50,256,4.0>

<1.10,256,39.5>

<1.40,-256,3.7>

<1.20,256,9.1> <0.90,256,11.8>

<1.20,256,16.2> <0.70,256,7.3>

<0.80,256,39.9>

<1.00,256,5.0> <1.00,256,17.9> <0.70,227,8.5>

<1.40,228,28.2>

<1.10,227,56.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-85
SLIDE 85

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<0.90,256,10.2> <0.40,-256,55.4>

<1.00,-256,10.7> <0.50,255,19.7> <1.40,252,74.6> <1.40,254,28.5> <1.10,256,58.2> <0.20,256,7.7> <1.00,256,7.1> <0.30,256,3.3> <1.60,256,7.4> <1.70,254,3.3> <1.10,256,4.2> <0.60,256,3.8> <1.10,256,3.5> <0.80,256,3.0> <2.00,256,3.4> <1.80,256,3.1> <0.80,256,3.1> <1.80,256,6.6> <0.10,256,4.8> <0.70,256,3.5>

<0.50,256,6.1>

<0.80,256,4.6>

<0.90,256,3.2> <0.90,256,7.0> <1.00,256,2.8> <1.50,256,5.3> <0.60,256,7.6>

<0.30,256,6.0>

<0.50,256,3.9> <0.40,256,20.5> <0.40,256,10.0> <0.80,256,4.0>

<1.00,256,39.3>

<1.40,-256,4.6>

<0.90,256,9.4> <0.90,256,10.3>

<0.60,256,16.0> <1.10,256,3.4>

<1.00,256,39.0>

<0.70,256,5.2> <1.00,256,32.5> <0.70,227,14.8>

<1.50,228,68.0>

<1.10,227,48.7>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-86
SLIDE 86

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<0.90,256,10.2> <0.40,-256,55.4>

<1.00,-256,10.7> <0.50,255,19.7> <1.40,252,74.6> <1.40,254,28.5> <1.10,256,58.2> <0.20,256,7.7> <1.00,256,7.1> <0.30,256,3.3> <1.60,256,7.4> <1.70,254,3.3> <1.10,256,4.2> <0.60,256,3.8> <1.10,256,3.5> <0.80,256,3.0> <2.00,256,3.4> <1.80,256,3.1> <1.80,256,6.6> <0.10,256,4.8> <0.70,256,3.5>

<0.50,256,6.1>

<0.80,256,4.6>

<0.90,256,3.2> <0.90,256,7.0> <1.00,256,2.8> <1.50,256,5.3> <0.60,256,7.6>

<0.30,256,6.0>

<0.50,256,3.9> <0.40,256,20.5> <0.40,256,10.0> <0.80,256,4.0>

<1.00,256,39.3>

<1.40,-256,4.6>

<0.90,256,9.4> <0.90,256,10.3>

<0.60,256,16.0> <1.10,256,3.4>

<1.00,256,39.0>

<0.70,256,5.2> <1.00,256,32.5> <0.70,227,14.8>

<1.50,228,68.0>

<1.10,227,48.7> <0.80,256,3.1> <0.80,256,3.1> <0.80,256,3.1> <0.80,256,3.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-87
SLIDE 87

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<0.90,256,10.2> <0.40,-256,55.4>

<1.00,-256,10.7> <0.50,255,19.7> <1.40,252,74.6> <1.40,254,28.5> <1.10,256,58.2> <0.20,256,7.7> <1.00,256,7.1> <0.30,256,3.3> <1.60,256,7.4> <1.70,254,3.3> <1.10,256,4.2> <0.60,256,3.8> <1.10,256,3.5> <0.80,256,3.0> <2.00,256,3.4> <1.80,256,3.1> <0.80,256,3.1> <0.80,256,3.1> <0.80,256,3.1> <0.80,256,3.1> <1.80,256,6.6> <0.10,256,4.8> <0.70,256,3.5>

<0.50,256,6.1>

<0.80,256,4.6>

<0.90,256,3.2> <0.90,256,7.0> <1.00,256,2.8> <1.50,256,5.3> <0.60,256,7.6>

<0.30,256,6.0>

<0.50,256,3.9> <0.40,256,20.5> <0.40,256,10.0> <0.80,256,4.0>

<1.00,256,39.3>

<1.40,-256,4.6>

<0.90,256,9.4> <0.90,256,10.3>

<0.60,256,16.0> <1.10,256,3.4>

<1.00,256,39.0>

<0.70,256,5.2> <1.00,256,32.5> <0.70,227,14.8>

<1.50,228,68.0>

<1.10,227,48.7>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-88
SLIDE 88

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.10,256,10.9> <0.40,-256,61.8>

<1.30,-256,5.0> <0.50,255,19.7> <1.40,252,75.3> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,8.2> <1.00,256,6.9> <0.30,256,3.1> <1.70,256,58.7> <1.50,254,4.2> <0.90,256,3.3> <0.20,256,3.2> <0.90,256,3.2> <0.40,256,3.1> <2.00,256,3.5> <1.90,256,3.3> <0.70,256,3.3> <0.80,256,4.6> <0.90,256,3.9> <0.80,256,3.2> <2.00,256,7.2> <0.30,256,4.1> <0.70,256,3.1>

<0.80,256,5.1>

<0.80,256,3.5>

<0.90,256,3.1> <0.90,256,7.1> <1.10,256,3.0> <1.40,256,4.7> <0.70,256,8.5>

<0.20,256,5.0>

<0.70,256,3.7> <0.20,256,14.8> <0.40,256,10.9> <0.90,256,5.4>

<0.70,256,39.8>

<1.70,-256,5.5>

<0.40,256,7.7> <0.90,256,12.6>

<0.80,256,9.5> <1.00,256,4.0>

<1.00,256,38.1>

<0.50,256,6.9> <1.00,256,34.1> <0.70,227,8.5>

<1.50,228,21.6>

<1.00,227,35.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-89
SLIDE 89

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.10,256,10.9> <0.40,-256,61.8>

<1.30,-256,5.0> <0.50,255,19.7> <1.40,252,75.3> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,8.2> <1.00,256,6.9> <0.30,256,3.1> <1.70,256,58.7> <1.50,254,4.2> <0.90,256,3.3> <0.20,256,3.2> <0.90,256,3.2> <0.40,256,3.1> <2.00,256,3.5> <1.90,256,3.3> <0.70,256,3.3> <0.80,256,4.6> <0.90,256,3.9> <0.80,256,3.2> <2.00,256,7.2> <0.30,256,4.1> <0.70,256,3.1>

<0.80,256,5.1>

<0.80,256,3.5>

<0.90,256,3.1> <0.90,256,7.1> <1.10,256,3.0> <1.40,256,4.7> <0.70,256,8.5>

<0.20,256,5.0>

<0.70,256,3.7> <0.20,256,14.8> <0.40,256,10.9>

<0.70,256,39.8>

<1.70,-256,5.5>

<0.40,256,7.7> <0.90,256,12.6>

<0.80,256,9.5> <1.00,256,4.0>

<1.00,256,38.1>

<0.50,256,6.9> <1.00,256,34.1> <0.70,227,8.5>

<1.50,228,21.6>

<1.00,227,35.8> <0.90,256,5.4> <0.90,256,5.4> <0.90,256,5.4> <0.90,256,5.4>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-90
SLIDE 90

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.10,256,10.9> <0.40,-256,61.8>

<1.30,-256,5.0> <0.50,255,19.7> <1.40,252,75.3> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,8.2> <1.00,256,6.9> <0.30,256,3.1> <1.70,256,58.7> <1.50,254,4.2> <0.90,256,3.3> <0.20,256,3.2> <0.90,256,3.2> <0.40,256,3.1> <2.00,256,3.5> <1.90,256,3.3> <0.70,256,3.3> <0.80,256,4.6> <0.90,256,3.9> <0.80,256,3.2> <2.00,256,7.2> <0.30,256,4.1> <0.70,256,3.1>

<0.80,256,5.1>

<0.80,256,3.5>

<0.90,256,3.1> <0.90,256,7.1> <1.10,256,3.0> <1.40,256,4.7> <0.70,256,8.5>

<0.20,256,5.0>

<0.70,256,3.7> <0.20,256,14.8> <0.40,256,10.9> <0.90,256,5.4> <0.90,256,5.4> <0.90,256,5.4> <0.90,256,5.4>

<0.70,256,39.8>

<1.70,-256,5.5>

<0.40,256,7.7> <0.90,256,12.6>

<0.80,256,9.5> <1.00,256,4.0>

<1.00,256,38.1>

<0.50,256,6.9> <1.00,256,34.1> <0.70,227,8.5>

<1.50,228,21.6>

<1.00,227,35.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-91
SLIDE 91

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.30,256,10.1> <0.40,-256,35.4>

<1.40,-256,50.7> <0.50,255,19.7> <1.40,252,24.2> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,14.7> <0.80,256,4.6> <0.30,256,4.0> <1.70,256,2.9> <1.70,254,3.5> <0.90,256,3.3> <0.30,256,4.8> <1.10,256,3.1> <0.80,256,3.2> <1.70,256,3.5> <1.90,256,3.9> <0.80,256,3.1> <0.30,256,3.0> <1.30,256,3.1> <0.70,256,3.3> <2.00,256,8.7> <0.20,256,3.9> <0.90,256,3.1>

<0.80,256,4.1>

<0.80,256,3.7>

<0.40,256,3.3> <0.90,256,6.4> <1.20,256,3.2> <1.10,256,4.7> <0.20,256,7.1>

<0.20,256,5.2>

<0.80,256,3.9> <0.40,256,14.8> <0.70,256,10.1> <0.80,256,6.1> <0.70,256,18.0> <0.90,256,6.1> <0.60,256,5.1>

<0.80,256,20.7>

<1.30,-256,5.4>

<0.70,256,7.0> <1.00,256,19.1>

<0.70,256,12.0> <1.10,256,11.1>

<1.20,256,40.8>

<0.30,256,7.7> <1.00,256,22.2> <0.60,227,51.8>

<1.70,228,10.3>

<1.00,227,29.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-92
SLIDE 92

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.30,256,10.1> <0.40,-256,35.4>

<1.40,-256,50.7> <0.50,255,19.7> <1.40,252,24.2> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,14.7> <0.80,256,4.6> <0.30,256,4.0> <1.70,256,2.9> <1.70,254,3.5> <0.90,256,3.3> <0.30,256,4.8> <1.10,256,3.1> <1.70,256,3.5> <1.90,256,3.9> <0.80,256,3.1> <0.30,256,3.0> <1.30,256,3.1> <0.70,256,3.3> <2.00,256,8.7> <0.20,256,3.9> <0.90,256,3.1>

<0.80,256,4.1>

<0.80,256,3.7>

<0.40,256,3.3> <0.90,256,6.4> <1.20,256,3.2> <1.10,256,4.7> <0.20,256,7.1>

<0.20,256,5.2>

<0.80,256,3.9> <0.40,256,14.8> <0.70,256,10.1> <0.80,256,6.1> <0.70,256,18.0> <0.90,256,6.1> <0.60,256,5.1>

<0.80,256,20.7>

<1.30,-256,5.4>

<0.70,256,7.0> <1.00,256,19.1>

<0.70,256,12.0> <1.10,256,11.1>

<1.20,256,40.8>

<0.30,256,7.7> <1.00,256,22.2> <0.60,227,51.8>

<1.70,228,10.3>

<1.00,227,29.9> <0.80,256,3.2> <0.80,256,3.2> <0.80,256,3.2> <0.80,256,3.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-93
SLIDE 93

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.30,256,10.1> <0.40,-256,35.4>

<1.40,-256,50.7> <0.50,255,19.7> <1.40,252,24.2> <1.40,254,22.2> <1.10,256,57.4> <0.20,256,14.7> <0.80,256,4.6> <0.30,256,4.0> <1.70,256,2.9> <1.70,254,3.5> <0.90,256,3.3> <0.30,256,4.8> <1.10,256,3.1> <0.80,256,3.2> <0.80,256,3.2> <0.80,256,3.2> <0.80,256,3.2> <1.70,256,3.5> <1.90,256,3.9> <0.80,256,3.1> <0.30,256,3.0> <1.30,256,3.1> <0.70,256,3.3> <2.00,256,8.7> <0.20,256,3.9> <0.90,256,3.1>

<0.80,256,4.1>

<0.80,256,3.7>

<0.40,256,3.3> <0.90,256,6.4> <1.20,256,3.2> <1.10,256,4.7> <0.20,256,7.1>

<0.20,256,5.2>

<0.80,256,3.9> <0.40,256,14.8> <0.70,256,10.1> <0.80,256,6.1> <0.70,256,18.0> <0.90,256,6.1> <0.60,256,5.1>

<0.80,256,20.7>

<1.30,-256,5.4>

<0.70,256,7.0> <1.00,256,19.1>

<0.70,256,12.0> <1.10,256,11.1>

<1.20,256,40.8>

<0.30,256,7.7> <1.00,256,22.2> <0.60,227,51.8>

<1.70,228,10.3>

<1.00,227,29.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-94
SLIDE 94

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.20,256,8.5> <0.60,-256,35.6>

<1.30,-256,46.9> <0.50,255,64.5> <1.40,252,24.2> <1.40,254,14.2> <1.10,256,3.5> <0.20,256,1.9> <0.70,256,4.4> <0.30,256,4.0> <1.70,256,10.9> <1.70,254,3.5> <0.90,256,4.1> <0.20,256,6.1> <1.10,256,2.3> <0.60,256,3.5> <0.80,256,3.8> <0.80,256,3.0> <0.80,256,3.6> <1.80,256,4.7> <1.80,256,3.2> <0.90,256,3.3> <0.30,256,3.6> <1.30,256,4.1> <0.80,256,3.5> <1.90,256,6.5> <0.20,256,3.2> <1.00,256,3.1>

<0.40,256,4.1>

<0.80,256,3.6>

<0.50,256,3.8> <1.20,256,6.5> <1.30,256,3.3> <1.10,256,5.4> <0.40,256,11.2>

<0.20,256,5.3>

<0.70,256,3.8> <0.30,256,13.9> <0.80,256,9.7> <0.60,256,5.8> <0.60,256,10.1> <0.80,256,5.1> <0.70,256,5.4>

<0.40,256,20.0>

<1.30,-256,6.3>

<0.70,256,6.7> <1.00,256,24.7>

<0.70,256,11.4> <0.90,256,17.6>

<1.30,256,35.4>

<0.60,256,8.8> <0.90,256,9.2> <0.60,227,46.2>

<1.70,228,15.7>

<1.00,227,37.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-95
SLIDE 95

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.20,256,8.5> <0.60,-256,35.6>

<1.30,-256,46.9> <0.50,255,64.5> <1.40,252,24.2> <1.40,254,14.2> <1.10,256,3.5> <0.20,256,1.9> <0.70,256,4.4> <0.30,256,4.0> <1.70,256,10.9> <1.70,254,3.5> <0.90,256,4.1> <0.20,256,6.1> <1.10,256,2.3> <0.60,256,3.5> <0.80,256,3.8> <0.80,256,3.0> <0.80,256,3.6> <1.80,256,4.7> <1.80,256,3.2> <0.90,256,3.3> <0.30,256,3.6> <1.30,256,4.1> <0.80,256,3.5> <1.90,256,6.5> <0.20,256,3.2> <1.00,256,3.1>

<0.40,256,4.1>

<0.80,256,3.6>

<0.50,256,3.8> <1.20,256,6.5> <1.30,256,3.3> <1.10,256,5.4> <0.40,256,11.2>

<0.20,256,5.3>

<0.30,256,13.9> <0.80,256,9.7> <0.60,256,5.8> <0.60,256,10.1> <0.80,256,5.1> <0.70,256,5.4>

<0.40,256,20.0>

<1.30,-256,6.3>

<0.70,256,6.7> <1.00,256,24.7>

<0.70,256,11.4> <0.90,256,17.6>

<1.30,256,35.4>

<0.60,256,8.8> <0.90,256,9.2> <0.60,227,46.2>

<1.70,228,15.7>

<1.00,227,37.1> <0.70,256,3.8> <0.70,256,3.8> <0.70,256,3.8> <0.70,256,3.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-96
SLIDE 96

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.20,256,8.5> <0.60,-256,35.6>

<1.30,-256,46.9> <0.50,255,64.5> <1.40,252,24.2> <1.40,254,14.2> <1.10,256,3.5> <0.20,256,1.9> <0.70,256,4.4> <0.30,256,4.0> <1.70,256,10.9> <1.70,254,3.5> <0.90,256,4.1> <0.20,256,6.1> <1.10,256,2.3> <0.60,256,3.5> <0.80,256,3.8> <0.80,256,3.0> <0.80,256,3.6> <1.80,256,4.7> <1.80,256,3.2> <0.90,256,3.3> <0.30,256,3.6> <1.30,256,4.1> <0.80,256,3.5> <1.90,256,6.5> <0.20,256,3.2> <1.00,256,3.1>

<0.40,256,4.1>

<0.80,256,3.6>

<0.50,256,3.8> <1.20,256,6.5> <1.30,256,3.3> <1.10,256,5.4> <0.40,256,11.2>

<0.20,256,5.3>

<0.70,256,3.8> <0.70,256,3.8> <0.70,256,3.8> <0.70,256,3.8> <0.30,256,13.9> <0.80,256,9.7> <0.60,256,5.8> <0.60,256,10.1> <0.80,256,5.1> <0.70,256,5.4>

<0.40,256,20.0>

<1.30,-256,6.3>

<0.70,256,6.7> <1.00,256,24.7>

<0.70,256,11.4> <0.90,256,17.6>

<1.30,256,35.4>

<0.60,256,8.8> <0.90,256,9.2> <0.60,227,46.2>

<1.70,228,15.7>

<1.00,227,37.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-97
SLIDE 97

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.30,256,8.5> <0.80,-256,23.6>

<1.20,-256,33.1> <0.50,255,63.8> <1.50,252,18.5> <1.40,254,8.6> <1.10,256,21.9> <0.10,256,1.1> <0.70,256,9.9> <0.40,256,10.4> <1.70,256,5.2> <1.70,254,3.5> <1.00,256,3.7> <0.30,256,8.0> <1.10,256,3.5> <0.60,256,3.4> <0.80,256,4.0> <0.90,256,3.1> <0.80,256,3.6> <1.80,256,4.0> <1.80,256,3.1> <0.80,256,3.4> <0.30,256,2.7> <1.20,256,3.1> <0.80,256,3.3> <1.90,256,4.8> <0.20,256,5.1> <1.00,256,3.1>

<0.40,256,5.5>

<1.00,256,3.2>

<0.60,256,3.9> <1.10,256,6.2> <1.20,256,5.7> <1.40,256,3.6> <0.30,256,12.0>

<0.40,256,4.1>

<1.20,256,3.5> <0.80,256,4.6> <0.70,256,4.0> <0.80,256,4.6> <0.20,256,13.6> <1.10,256,11.5> <0.60,256,4.9> <0.70,256,11.7> <0.80,256,11.4> <0.70,256,5.4>

<0.50,256,20.7>

<1.60,-256,5.7>

<0.50,256,7.3> <0.80,256,24.8>

<0.60,256,10.5> <0.70,256,11.2>

<1.30,256,41.9>

<0.50,256,7.1> <0.90,256,10.0> <0.60,227,60.8>

<1.70,228,16.5>

<1.00,227,57.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-98
SLIDE 98

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.30,256,8.5> <0.80,-256,23.6>

<1.20,-256,33.1> <0.50,255,63.8> <1.50,252,18.5> <1.40,254,8.6> <1.10,256,21.9> <0.10,256,1.1> <0.70,256,9.9> <0.40,256,10.4> <1.70,256,5.2> <1.70,254,3.5> <1.00,256,3.7> <0.30,256,8.0> <1.10,256,3.5> <0.60,256,3.4> <0.80,256,4.0> <0.90,256,3.1> <0.80,256,3.6> <1.80,256,4.0> <1.80,256,3.1> <0.80,256,3.4> <0.30,256,2.7> <1.20,256,3.1> <0.80,256,3.3> <1.90,256,4.8> <0.20,256,5.1> <1.00,256,3.1>

<0.40,256,5.5>

<1.00,256,3.2>

<0.60,256,3.9> <1.10,256,6.2> <1.20,256,5.7> <1.40,256,3.6> <0.30,256,12.0>

<0.40,256,4.1>

<1.20,256,3.5> <0.80,256,4.6> <0.70,256,4.0> <0.80,256,4.6> <0.20,256,13.6> <1.10,256,11.5> <0.60,256,4.9> <0.70,256,11.7> <0.80,256,11.4> <0.70,256,5.4>

<0.50,256,20.7>

<1.60,-256,5.7>

<0.80,256,24.8>

<0.60,256,10.5> <0.70,256,11.2>

<1.30,256,41.9>

<0.50,256,7.1> <0.90,256,10.0> <0.60,227,60.8>

<1.70,228,16.5>

<1.00,227,57.1> <0.50,256,7.3> <0.50,256,7.3> <0.50,256,7.3> <0.50,256,7.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-99
SLIDE 99

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.30,256,8.5> <0.80,-256,23.6>

<1.20,-256,33.1> <0.50,255,63.8> <1.50,252,18.5> <1.40,254,8.6> <1.10,256,21.9> <0.10,256,1.1> <0.70,256,9.9> <0.40,256,10.4> <1.70,256,5.2> <1.70,254,3.5> <1.00,256,3.7> <0.30,256,8.0> <1.10,256,3.5> <0.60,256,3.4> <0.80,256,4.0> <0.90,256,3.1> <0.80,256,3.6> <1.80,256,4.0> <1.80,256,3.1> <0.80,256,3.4> <0.30,256,2.7> <1.20,256,3.1> <0.80,256,3.3> <1.90,256,4.8> <0.20,256,5.1> <1.00,256,3.1>

<0.40,256,5.5>

<1.00,256,3.2>

<0.60,256,3.9> <1.10,256,6.2> <1.20,256,5.7> <1.40,256,3.6> <0.30,256,12.0>

<0.40,256,4.1>

<1.20,256,3.5> <0.80,256,4.6> <0.70,256,4.0> <0.80,256,4.6> <0.20,256,13.6> <1.10,256,11.5> <0.60,256,4.9> <0.70,256,11.7> <0.80,256,11.4> <0.70,256,5.4>

<0.50,256,20.7>

<1.60,-256,5.7>

<0.50,256,7.3> <0.50,256,7.3> <0.50,256,7.3> <0.50,256,7.3>

<0.80,256,24.8>

<0.60,256,10.5> <0.70,256,11.2>

<1.30,256,41.9>

<0.50,256,7.1> <0.90,256,10.0> <0.60,227,60.8>

<1.70,228,16.5>

<1.00,227,57.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-100
SLIDE 100

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.40,256,8.3> <0.70,-256,29.0>

<1.40,-256,27.5> <0.50,255,69.4> <1.50,252,18.5> <1.40,254,14.3> <1.10,256,73.9> <0.20,256,58.7> <0.90,256,2.8> <0.50,256,3.3> <1.80,256,2.8> <1.70,254,3.2> <1.00,256,3.8> <0.30,256,9.8> <1.00,256,5.9> <0.50,256,3.4> <0.90,256,3.4> <1.00,256,2.8> <0.80,256,2.9> <1.50,256,3.1> <1.90,256,3.1> <0.60,256,3.3> <0.50,256,4.4> <1.20,256,3.1> <0.80,256,4.0> <1.90,256,5.6> <0.20,256,3.5> <1.10,256,3.2>

<0.40,256,4.9>

<1.00,256,3.0>

<0.90,256,3.6> <1.00,256,5.6> <1.40,256,5.0> <1.30,256,3.5> <0.40,256,18.3>

<0.30,256,3.8>

<1.20,256,3.9> <0.50,256,12.1> <0.80,256,5.1> <0.80,256,4.9> <0.20,256,13.7> <1.30,256,9.9> <0.80,256,4.1> <1.00,256,11.3> <1.30,256,17.9> <0.70,256,4.6>

<0.30,256,21.5>

<2.00,-256,4.2>

<0.40,256,6.6> <0.50,256,7.2> <0.90,256,8.7> <0.80,256,8.0>

<0.80,256,19.1>

<0.50,256,11.1> <0.80,256,9.8>

<1.20,256,40.9>

<0.50,256,8.6> <0.90,256,10.3> <0.60,227,21.6>

<1.70,228,61.2>

<1.00,227,57.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-101
SLIDE 101

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.40,256,8.3> <0.70,-256,29.0>

<1.40,-256,27.5> <0.50,255,69.4> <1.50,252,18.5> <1.40,254,14.3> <1.10,256,73.9> <0.20,256,58.7> <0.90,256,2.8> <0.50,256,3.3> <1.80,256,2.8> <1.70,254,3.2> <1.00,256,3.8> <0.30,256,9.8> <1.00,256,5.9> <0.50,256,3.4> <0.90,256,3.4> <1.00,256,2.8> <0.80,256,2.9> <1.50,256,3.1> <1.90,256,3.1> <0.60,256,3.3> <0.50,256,4.4> <1.20,256,3.1> <0.80,256,4.0> <1.90,256,5.6> <0.20,256,3.5>

<0.40,256,4.9>

<1.00,256,3.0>

<0.90,256,3.6> <1.00,256,5.6> <1.40,256,5.0> <1.30,256,3.5> <0.40,256,18.3>

<0.30,256,3.8>

<1.20,256,3.9> <0.50,256,12.1> <0.80,256,5.1> <0.80,256,4.9> <0.20,256,13.7> <1.30,256,9.9> <0.80,256,4.1> <1.00,256,11.3> <1.30,256,17.9> <0.70,256,4.6>

<0.30,256,21.5>

<2.00,-256,4.2>

<0.40,256,6.6> <0.50,256,7.2> <0.90,256,8.7> <0.80,256,8.0>

<0.80,256,19.1>

<0.50,256,11.1> <0.80,256,9.8>

<1.20,256,40.9>

<0.50,256,8.6> <0.90,256,10.3> <0.60,227,21.6>

<1.70,228,61.2>

<1.00,227,57.1> <1.10,256,3.2> <1.10,256,3.2> <1.10,256,3.2> <1.10,256,3.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-102
SLIDE 102

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.40,256,8.3> <0.70,-256,29.0>

<1.40,-256,27.5> <0.50,255,69.4> <1.50,252,18.5> <1.40,254,14.3> <1.10,256,73.9> <0.20,256,58.7> <0.90,256,2.8> <0.50,256,3.3> <1.80,256,2.8> <1.70,254,3.2> <1.00,256,3.8> <0.30,256,9.8> <1.00,256,5.9> <0.50,256,3.4> <0.90,256,3.4> <1.00,256,2.8> <0.80,256,2.9> <1.50,256,3.1> <1.90,256,3.1> <0.60,256,3.3> <0.50,256,4.4> <1.20,256,3.1> <0.80,256,4.0> <1.90,256,5.6> <0.20,256,3.5> <1.10,256,3.2> <1.10,256,3.2> <1.10,256,3.2> <1.10,256,3.2>

<0.40,256,4.9>

<1.00,256,3.0>

<0.90,256,3.6> <1.00,256,5.6> <1.40,256,5.0> <1.30,256,3.5> <0.40,256,18.3>

<0.30,256,3.8>

<1.20,256,3.9> <0.50,256,12.1> <0.80,256,5.1> <0.80,256,4.9> <0.20,256,13.7> <1.30,256,9.9> <0.80,256,4.1> <1.00,256,11.3> <1.30,256,17.9> <0.70,256,4.6>

<0.30,256,21.5>

<2.00,-256,4.2>

<0.40,256,6.6> <0.50,256,7.2> <0.90,256,8.7> <0.80,256,8.0>

<0.80,256,19.1>

<0.50,256,11.1> <0.80,256,9.8>

<1.20,256,40.9>

<0.50,256,8.6> <0.90,256,10.3> <0.60,227,21.6>

<1.70,228,61.2>

<1.00,227,57.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-103
SLIDE 103

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.90,256,9.4> <0.80,-256,26.5>

<1.40,-256,72.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,65.7> <1.10,256,9.9> <0.20,256,6.8> <0.80,256,16.3> <0.50,256,3.0> <1.80,256,16.2> <1.80,254,3.1> <1.00,256,3.6> <0.30,256,2.8> <1.00,256,5.8> <0.50,256,2.5> <0.90,256,8.9> <0.90,256,2.9> <0.80,256,3.8> <1.30,256,3.5> <2.00,256,4.1> <1.00,256,3.0> <0.50,256,3.3> <1.30,256,3.7> <0.90,256,3.8> <2.00,256,5.5> <0.20,256,3.4> <1.10,256,3.0> <1.20,256,4.8> <1.10,256,4.8> <1.00,256,3.3>

<0.40,256,5.1>

<0.80,256,3.6>

<0.50,256,3.6> <1.10,256,4.9> <1.40,256,5.8> <1.30,256,3.5> <0.40,256,17.4>

<0.20,256,3.1>

<1.10,256,3.9> <0.40,256,18.5> <0.70,256,11.5> <0.70,256,3.3> <0.30,256,11.3> <1.40,256,8.5> <1.00,256,4.8> <1.00,256,12.5> <1.50,256,7.4> <0.70,256,5.6>

<0.30,256,22.4>

<1.90,-256,5.1>

<0.40,256,6.7> <0.40,256,13.6> <0.60,256,9.0> <1.10,256,8.0>

<1.20,256,18.4>

<0.40,256,15.8> <0.90,256,8.8>

<1.10,256,39.8>

<0.40,256,10.1> <0.80,256,8.8> <0.60,227,72.7>

<1.70,228,9.0>

<1.00,227,51.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-104
SLIDE 104

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.90,256,9.4> <0.80,-256,26.5>

<1.40,-256,72.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,65.7> <1.10,256,9.9> <0.20,256,6.8> <0.80,256,16.3> <0.50,256,3.0> <1.80,256,16.2> <1.80,254,3.1> <1.00,256,3.6> <0.30,256,2.8> <1.00,256,5.8> <0.50,256,2.5> <0.90,256,8.9> <0.90,256,2.9> <0.80,256,3.8> <1.30,256,3.5> <2.00,256,4.1> <1.00,256,3.0> <0.50,256,3.3> <1.30,256,3.7> <0.90,256,3.8> <2.00,256,5.5> <0.20,256,3.4> <1.10,256,3.0> <1.20,256,4.8> <1.10,256,4.8> <1.00,256,3.3>

<0.40,256,5.1>

<0.80,256,3.6>

<0.50,256,3.6> <1.10,256,4.9> <1.40,256,5.8> <1.30,256,3.5> <0.40,256,17.4>

<0.20,256,3.1>

<1.10,256,3.9> <0.40,256,18.5> <0.70,256,11.5> <0.70,256,3.3> <0.30,256,11.3> <1.40,256,8.5> <1.00,256,4.8> <1.00,256,12.5> <1.50,256,7.4> <0.70,256,5.6>

<0.30,256,22.4>

<1.90,-256,5.1>

<0.40,256,6.7> <0.40,256,13.6> <0.60,256,9.0>

<1.20,256,18.4>

<0.40,256,15.8> <0.90,256,8.8>

<1.10,256,39.8>

<0.40,256,10.1> <0.80,256,8.8> <0.60,227,72.7>

<1.70,228,9.0>

<1.00,227,51.8> <1.10,256,8.0> <1.10,256,8.0> <1.10,256,8.0> <1.10,256,8.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-105
SLIDE 105

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.90,256,9.4> <0.80,-256,26.5>

<1.40,-256,72.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,65.7> <1.10,256,9.9> <0.20,256,6.8> <0.80,256,16.3> <0.50,256,3.0> <1.80,256,16.2> <1.80,254,3.1> <1.00,256,3.6> <0.30,256,2.8> <1.00,256,5.8> <0.50,256,2.5> <0.90,256,8.9> <0.90,256,2.9> <0.80,256,3.8> <1.30,256,3.5> <2.00,256,4.1> <1.00,256,3.0> <0.50,256,3.3> <1.30,256,3.7> <0.90,256,3.8> <2.00,256,5.5> <0.20,256,3.4> <1.10,256,3.0> <1.20,256,4.8> <1.10,256,4.8> <1.00,256,3.3>

<0.40,256,5.1>

<0.80,256,3.6>

<0.50,256,3.6> <1.10,256,4.9> <1.40,256,5.8> <1.30,256,3.5> <0.40,256,17.4>

<0.20,256,3.1>

<1.10,256,3.9> <0.40,256,18.5> <0.70,256,11.5> <0.70,256,3.3> <0.30,256,11.3> <1.40,256,8.5> <1.00,256,4.8> <1.00,256,12.5> <1.50,256,7.4> <0.70,256,5.6>

<0.30,256,22.4>

<1.90,-256,5.1>

<0.40,256,6.7> <0.40,256,13.6> <0.60,256,9.0> <1.10,256,8.0> <1.10,256,8.0> <1.10,256,8.0> <1.10,256,8.0>

<1.20,256,18.4>

<0.40,256,15.8> <0.90,256,8.8>

<1.10,256,39.8>

<0.40,256,10.1> <0.80,256,8.8> <0.60,227,72.7>

<1.70,228,9.0>

<1.00,227,51.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-106
SLIDE 106

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<1.90,256,9.8> <0.70,-256,11.3>

<1.50,-256,28.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,59.3> <1.10,256,4.3> <0.20,256,65.8> <0.70,256,10.1> <0.40,256,3.2> <1.80,256,5.8> <1.70,254,2.8> <1.00,256,3.1> <0.30,256,3.5> <1.00,256,4.1> <0.50,256,3.3> <0.90,256,8.1> <0.90,256,4.3> <0.70,256,2.9> <1.40,256,3.9> <2.00,256,3.3> <1.00,256,3.6> <0.50,256,4.2> <1.30,256,3.0> <0.80,256,3.0> <2.00,256,4.7> <0.30,256,3.5> <1.20,256,3.2> <1.30,256,5.0> <0.80,256,3.9> <1.00,256,3.0>

<0.50,256,5.8>

<1.10,256,3.3>

<0.60,256,3.3> <1.20,256,4.2> <1.50,256,6.7> <1.30,256,3.5> <0.40,256,23.1>

<0.10,256,3.2>

<1.10,256,3.9> <0.40,256,16.9> <0.90,256,4.9> <1.00,256,3.6> <0.20,256,10.2> <1.50,256,6.0> <0.90,256,5.0> <0.90,256,12.6> <1.40,256,5.9> <1.20,256,6.0>

<0.10,256,16.1>

<1.70,-256,5.1>

<0.10,256,6.9> <0.50,256,15.3> <0.70,256,9.6> <1.50,256,8.1> <1.40,256,10.3> <1.20,256,7.9> <1.30,256,7.9>

<1.20,256,19.2>

<0.50,256,15.1> <1.30,256,9.2>

<1.00,256,33.4>

<0.80,256,10.6> <0.90,256,6.4> <0.50,227,67.0>

<1.70,228,7.9>

<1.10,227,58.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-107
SLIDE 107

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<1.90,256,9.8> <0.70,-256,11.3>

<1.50,-256,28.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,59.3> <1.10,256,4.3> <0.20,256,65.8> <0.70,256,10.1> <0.40,256,3.2> <1.80,256,5.8> <1.70,254,2.8> <1.00,256,3.1> <0.30,256,3.5> <1.00,256,4.1> <0.50,256,3.3> <0.90,256,8.1> <0.90,256,4.3> <0.70,256,2.9> <1.40,256,3.9> <2.00,256,3.3> <1.00,256,3.6> <0.50,256,4.2> <1.30,256,3.0> <0.80,256,3.0> <2.00,256,4.7> <0.30,256,3.5> <1.20,256,3.2> <1.30,256,5.0> <0.80,256,3.9> <1.00,256,3.0>

<0.50,256,5.8>

<1.10,256,3.3>

<1.20,256,4.2> <1.50,256,6.7> <1.30,256,3.5> <0.40,256,23.1>

<0.10,256,3.2>

<1.10,256,3.9> <0.40,256,16.9> <0.90,256,4.9> <1.00,256,3.6> <0.20,256,10.2> <1.50,256,6.0> <0.90,256,5.0> <0.90,256,12.6> <1.40,256,5.9> <1.20,256,6.0>

<0.10,256,16.1>

<1.70,-256,5.1>

<0.10,256,6.9> <0.50,256,15.3> <0.70,256,9.6> <1.50,256,8.1> <1.40,256,10.3> <1.20,256,7.9> <1.30,256,7.9>

<1.20,256,19.2>

<0.50,256,15.1> <1.30,256,9.2>

<1.00,256,33.4>

<0.80,256,10.6> <0.90,256,6.4> <0.50,227,67.0>

<1.70,228,7.9>

<1.10,227,58.2> <0.60,256,3.3> <0.60,256,3.3> <0.60,256,3.3> <0.60,256,3.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-108
SLIDE 108

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<1.90,256,9.8> <0.70,-256,11.3>

<1.50,-256,28.9> <0.50,255,71.0> <1.60,252,31.4> <1.40,254,59.3> <1.10,256,4.3> <0.20,256,65.8> <0.70,256,10.1> <0.40,256,3.2> <1.80,256,5.8> <1.70,254,2.8> <1.00,256,3.1> <0.30,256,3.5> <1.00,256,4.1> <0.50,256,3.3> <0.90,256,8.1> <0.90,256,4.3> <0.70,256,2.9> <1.40,256,3.9> <2.00,256,3.3> <1.00,256,3.6> <0.50,256,4.2> <1.30,256,3.0> <0.80,256,3.0> <2.00,256,4.7> <0.30,256,3.5> <1.20,256,3.2> <1.30,256,5.0> <0.80,256,3.9> <1.00,256,3.0>

<0.50,256,5.8>

<1.10,256,3.3>

<0.60,256,3.3> <0.60,256,3.3> <0.60,256,3.3> <0.60,256,3.3> <1.20,256,4.2> <1.50,256,6.7> <1.30,256,3.5> <0.40,256,23.1>

<0.10,256,3.2>

<1.10,256,3.9> <0.40,256,16.9> <0.90,256,4.9> <1.00,256,3.6> <0.20,256,10.2> <1.50,256,6.0> <0.90,256,5.0> <0.90,256,12.6> <1.40,256,5.9> <1.20,256,6.0>

<0.10,256,16.1>

<1.70,-256,5.1>

<0.10,256,6.9> <0.50,256,15.3> <0.70,256,9.6> <1.50,256,8.1> <1.40,256,10.3> <1.20,256,7.9> <1.30,256,7.9>

<1.20,256,19.2>

<0.50,256,15.1> <1.30,256,9.2>

<1.00,256,33.4>

<0.80,256,10.6> <0.90,256,6.4> <0.50,227,67.0>

<1.70,228,7.9>

<1.10,227,58.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-109
SLIDE 109

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<2.00,256,10.4> <0.80,-256,51.3>

<1.30,-256,29.7> <0.50,255,0.6> <1.60,252,19.4> <1.40,254,66.5> <1.10,256,3.9> <0.20,256,21.1> <0.70,256,3.7> <0.40,256,3.9> <1.90,256,3.3> <1.50,254,3.0> <0.90,256,3.3> <0.20,256,3.5> <0.90,256,3.0> <0.50,256,3.3> <1.00,256,5.8> <0.90,256,4.3> <0.70,256,2.8> <1.30,256,3.1> <1.70,256,3.2> <1.10,256,3.2> <0.40,256,3.3> <1.60,256,3.4> <0.80,256,3.1> <2.00,256,5.9> <0.30,256,4.1> <1.10,256,3.4> <1.50,256,3.5> <0.90,256,3.3> <0.90,256,2.7>

<0.70,256,6.0>

<0.90,256,3.3>

<0.60,256,3.6> <1.00,256,3.0> <0.80,256,4.8> <0.70,256,4.2> <1.60,256,4.0> <1.50,256,7.7> <1.10,256,3.5> <0.40,256,23.0>

<0.20,256,3.7>

<1.00,256,3.9> <0.50,256,9.6> <0.80,256,5.1> <1.30,256,3.4> <0.70,256,7.8> <1.50,256,5.2> <1.10,256,5.1> <0.60,256,13.5> <1.30,256,14.7> <1.10,256,5.3>

<0.10,256,13.1>

<1.80,-256,6.6>

<0.20,256,7.2> <0.10,256,20.0> <0.70,256,8.8> <1.10,256,8.1> <1.60,256,9.9> <1.20,256,7.9> <1.50,256,7.9>

<1.30,256,18.4>

<0.30,256,14.2> <1.30,256,9.2>

<1.00,256,38.5>

<1.20,256,15.4> <0.80,256,7.4> <0.50,227,22.0>

<2.00,228,8.7>

<1.00,227,38.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-110
SLIDE 110

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<2.00,256,10.4> <0.80,-256,51.3>

<1.30,-256,29.7> <0.50,255,0.6> <1.60,252,19.4> <1.40,254,66.5> <1.10,256,3.9> <0.20,256,21.1> <0.70,256,3.7> <0.40,256,3.9> <1.90,256,3.3> <1.50,254,3.0> <0.90,256,3.3> <0.20,256,3.5> <0.90,256,3.0> <0.50,256,3.3> <1.00,256,5.8> <0.90,256,4.3> <0.70,256,2.8> <1.30,256,3.1> <1.70,256,3.2> <1.10,256,3.2> <0.40,256,3.3> <1.60,256,3.4> <0.80,256,3.1> <2.00,256,5.9> <0.30,256,4.1> <1.10,256,3.4> <1.50,256,3.5> <0.90,256,3.3> <0.90,256,2.7>

<0.70,256,6.0>

<0.90,256,3.3>

<0.60,256,3.6> <1.00,256,3.0> <0.80,256,4.8> <0.70,256,4.2> <1.60,256,4.0> <1.50,256,7.7> <1.10,256,3.5> <0.40,256,23.0>

<0.20,256,3.7>

<1.00,256,3.9> <0.50,256,9.6> <0.80,256,5.1> <1.30,256,3.4> <0.70,256,7.8> <1.50,256,5.2> <1.10,256,5.1> <0.60,256,13.5> <1.30,256,14.7> <1.10,256,5.3>

<0.10,256,13.1>

<1.80,-256,6.6>

<0.20,256,7.2> <0.10,256,20.0> <0.70,256,8.8> <1.60,256,9.9> <1.20,256,7.9> <1.50,256,7.9>

<1.30,256,18.4>

<0.30,256,14.2> <1.30,256,9.2>

<1.00,256,38.5>

<1.20,256,15.4> <0.80,256,7.4> <0.50,227,22.0>

<2.00,228,8.7>

<1.00,227,38.8> <1.10,256,8.1> <1.10,256,8.1> <1.10,256,8.1> <1.10,256,8.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-111
SLIDE 111

Introduction The problem Remy Evaluation Discussion

Optimize

s_ewma r_ewma

<2.00,256,10.4> <0.80,-256,51.3>

<1.30,-256,29.7> <0.50,255,0.6> <1.60,252,19.4> <1.40,254,66.5> <1.10,256,3.9> <0.20,256,21.1> <0.70,256,3.7> <0.40,256,3.9> <1.90,256,3.3> <1.50,254,3.0> <0.90,256,3.3> <0.20,256,3.5> <0.90,256,3.0> <0.50,256,3.3> <1.00,256,5.8> <0.90,256,4.3> <0.70,256,2.8> <1.30,256,3.1> <1.70,256,3.2> <1.10,256,3.2> <0.40,256,3.3> <1.60,256,3.4> <0.80,256,3.1> <2.00,256,5.9> <0.30,256,4.1> <1.10,256,3.4> <1.50,256,3.5> <0.90,256,3.3> <0.90,256,2.7>

<0.70,256,6.0>

<0.90,256,3.3>

<0.60,256,3.6> <1.00,256,3.0> <0.80,256,4.8> <0.70,256,4.2> <1.60,256,4.0> <1.50,256,7.7> <1.10,256,3.5> <0.40,256,23.0>

<0.20,256,3.7>

<1.00,256,3.9> <0.50,256,9.6> <0.80,256,5.1> <1.30,256,3.4> <0.70,256,7.8> <1.50,256,5.2> <1.10,256,5.1> <0.60,256,13.5> <1.30,256,14.7> <1.10,256,5.3>

<0.10,256,13.1>

<1.80,-256,6.6>

<0.20,256,7.2> <0.10,256,20.0> <0.70,256,8.8> <1.10,256,8.1> <1.10,256,8.1> <1.10,256,8.1> <1.10,256,8.1> <1.60,256,9.9> <1.20,256,7.9> <1.50,256,7.9>

<1.30,256,18.4>

<0.30,256,14.2> <1.30,256,9.2>

<1.00,256,38.5>

<1.20,256,15.4> <0.80,256,7.4> <0.50,227,22.0>

<2.00,228,8.7>

<1.00,227,38.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-112
SLIDE 112

Introduction The problem Remy Evaluation Discussion

Split

s_ewma r_ewma

<2.00,256,9.7> <0.70,-256,50.2>

<1.50,-256,41.6> <0.50,255,60.6> <1.60,252,12.1> <1.40,254,72.3> <1.10,256,4.6> <0.20,256,14.1> <0.70,256,3.8> <0.40,256,3.1> <1.90,256,2.7> <1.60,254,3.0> <0.90,256,3.1> <0.20,256,6.2> <0.90,256,4.0> <0.50,256,3.2> <1.00,256,5.1> <0.90,256,2.7> <0.70,256,3.0> <1.20,256,3.7> <1.70,256,3.1> <1.00,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.1> <1.90,256,5.2> <0.60,256,3.7> <1.10,256,3.5> <1.50,256,5.1> <0.80,256,3.3> <1.00,256,2.7>

<0.90,256,6.8>

<0.60,256,3.6>

<0.70,256,5.5> <1.10,256,3.1> <0.70,256,3.9> <1.00,256,3.2> <1.30,256,4.3> <1.50,256,4.5> <1.50,256,3.6> <0.20,256,10.0>

<0.20,256,3.6>

<0.90,256,4.5> <0.70,256,9.8> <0.40,256,6.0> <1.50,256,3.0> <0.50,256,6.1> <1.60,256,6.1> <0.90,256,4.9> <0.70,256,14.3> <1.10,256,14.5> <1.30,256,5.4>

<0.20,256,12.3>

<1.70,-256,7.7>

<0.20,256,7.0> <0.20,256,24.7> <1.10,256,5.6> <0.80,256,6.9> <0.80,256,21.4> <1.30,256,13.7> <1.30,256,7.2> <1.60,256,10.6> <1.30,256,14.2> <1.40,256,7.9>

<1.30,256,18.3>

<0.50,256,18.4> <1.20,256,8.5>

<1.30,256,25.8>

<1.10,256,9.8> <0.50,256,8.9> <0.70,227,59.8>

<1.80,228,8.8>

<1.20,227,38.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-113
SLIDE 113

Introduction The problem Remy Evaluation Discussion

Simulate

s_ewma r_ewma

<2.00,256,9.7> <0.70,-256,50.2>

<1.50,-256,41.6> <0.50,255,60.6> <1.60,252,12.1> <1.40,254,72.3> <1.10,256,4.6> <0.20,256,14.1> <0.70,256,3.8> <0.40,256,3.1> <1.90,256,2.7> <1.60,254,3.0> <0.90,256,3.1> <0.20,256,6.2> <0.90,256,4.0> <0.50,256,3.2> <1.00,256,5.1> <0.90,256,2.7> <0.70,256,3.0> <1.20,256,3.7> <1.70,256,3.1> <1.00,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.1> <1.90,256,5.2> <0.60,256,3.7> <1.10,256,3.5> <1.50,256,5.1> <0.80,256,3.3> <1.00,256,2.7>

<0.90,256,6.8>

<0.60,256,3.6>

<0.70,256,5.5> <1.10,256,3.1> <0.70,256,3.9> <1.00,256,3.2> <1.30,256,4.3> <1.50,256,4.5> <1.50,256,3.6> <0.20,256,10.0>

<0.20,256,3.6>

<0.90,256,4.5> <0.70,256,9.8> <0.40,256,6.0> <1.50,256,3.0> <0.50,256,6.1> <1.60,256,6.1> <0.90,256,4.9> <0.70,256,14.3> <1.10,256,14.5> <1.30,256,5.4>

<0.20,256,12.3>

<1.70,-256,7.7>

<0.20,256,7.0> <0.20,256,24.7> <1.10,256,5.6> <0.80,256,6.9> <0.80,256,21.4> <1.30,256,13.7> <1.30,256,7.2> <1.60,256,10.6> <1.30,256,14.2>

<1.30,256,18.3>

<0.50,256,18.4> <1.20,256,8.5>

<1.30,256,25.8>

<1.10,256,9.8> <0.50,256,8.9> <0.70,227,59.8>

<1.80,228,8.8>

<1.20,227,38.1> <1.40,256,7.9> <1.40,256,7.9> <1.40,256,7.9> <1.40,256,7.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-114
SLIDE 114

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<2.00,256,9.7> <0.70,-256,50.2>

<1.50,-256,41.6> <0.50,255,60.6> <1.60,252,12.1> <1.40,254,72.3> <1.10,256,4.6> <0.20,256,14.1> <0.70,256,3.8> <0.40,256,3.1> <1.90,256,2.7> <1.60,254,3.0> <0.90,256,3.1> <0.20,256,6.2> <0.90,256,4.0> <0.50,256,3.2> <1.00,256,5.1> <0.90,256,2.7> <0.70,256,3.0> <1.20,256,3.7> <1.70,256,3.1> <1.00,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.1> <1.90,256,5.2> <0.60,256,3.7> <1.10,256,3.5> <1.50,256,5.1> <0.80,256,3.3> <1.00,256,2.7>

<0.90,256,6.8>

<0.60,256,3.6>

<0.70,256,5.5> <1.10,256,3.1> <0.70,256,3.9> <1.00,256,3.2> <1.30,256,4.3> <1.50,256,4.5> <1.50,256,3.6> <0.20,256,10.0>

<0.20,256,3.6>

<0.90,256,4.5> <0.70,256,9.8> <0.40,256,6.0> <1.50,256,3.0> <0.50,256,6.1> <1.60,256,6.1> <0.90,256,4.9> <0.70,256,14.3> <1.10,256,14.5> <1.30,256,5.4>

<0.20,256,12.3>

<1.70,-256,7.7>

<0.20,256,7.0> <0.20,256,24.7> <1.10,256,5.6> <0.80,256,6.9> <0.80,256,21.4> <1.30,256,13.7> <1.30,256,7.2> <1.60,256,10.6> <1.30,256,14.2> <1.40,256,7.9> <1.40,256,7.9> <1.40,256,7.9> <1.40,256,7.9>

<1.30,256,18.3>

<0.50,256,18.4> <1.20,256,8.5>

<1.30,256,25.8>

<1.10,256,9.8> <0.50,256,8.9> <0.70,227,59.8>

<1.80,228,8.8>

<1.20,227,38.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-115
SLIDE 115

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.60,256,9.6> <1.00,-256,43.1>

<1.60,-256,55.1> <0.50,255,10.8> <1.60,252,63.2> <1.40,254,20.3> <1.10,256,3.4> <0.20,256,2.8> <0.60,256,3.0> <0.40,256,3.1> <2.00,256,4.0> <1.50,254,3.3> <0.90,256,2.8> <0.40,256,4.9> <0.90,256,3.0> <0.60,256,3.0> <1.00,256,6.7> <1.10,256,2.9> <0.70,256,3.1> <1.00,256,4.6> <1.80,256,3.4> <0.70,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.0> <1.70,256,3.9> <0.70,256,4.4> <1.20,256,3.2> <1.70,256,4.3> <0.90,256,3.3> <1.10,256,2.5>

<1.00,256,6.3>

<0.60,256,3.4>

<0.50,256,5.3> <1.20,256,4.6> <0.40,256,4.0> <0.80,256,3.1> <1.50,256,4.9> <1.80,256,5.4> <1.50,256,4.1> <0.20,256,9.8>

<0.30,256,4.6>

<1.40,256,2.9> <0.90,256,10.6> <0.60,256,12.5> <1.60,256,4.5> <0.50,256,6.9> <1.30,256,3.4> <1.00,256,5.2> <0.20,256,21.8> <1.20,256,13.7> <1.20,256,4.7>

<0.40,256,25.8>

<2.00,-256,6.8>

<0.20,256,5.5> <0.90,256,12.7> <1.00,256,7.1> <0.60,256,8.4> <0.30,256,13.4> <1.50,256,12.2> <1.30,256,8.1> <1.60,256,8.9> <1.00,256,8.6> <1.50,256,8.0> <1.50,256,13.4> <1.60,256,9.2> <1.30,256,8.2>

<1.00,256,17.0>

<0.50,256,12.0> <1.00,256,8.7>

<1.50,256,30.7>

<1.50,256,9.6> <0.80,256,11.2> <0.80,227,10.4>

<1.30,228,7.2>

<1.10,227,43.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-116
SLIDE 116

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.60,256,9.6> <1.00,-256,43.1>

<1.60,-256,55.1> <0.50,255,10.8> <1.60,252,63.2> <1.40,254,20.3> <1.10,256,3.4> <0.20,256,2.8> <0.60,256,3.0> <0.40,256,3.1> <2.00,256,4.0> <1.50,254,3.3> <0.90,256,2.8> <0.40,256,4.9> <0.90,256,3.0> <0.60,256,3.0> <1.00,256,6.7> <1.10,256,2.9> <0.70,256,3.1> <1.00,256,4.6> <1.80,256,3.4> <0.70,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.0> <1.70,256,3.9> <0.70,256,4.4> <1.20,256,3.2> <1.70,256,4.3> <0.90,256,3.3> <1.10,256,2.5>

<1.00,256,6.3>

<0.60,256,3.4>

<0.50,256,5.3> <1.20,256,4.6> <0.40,256,4.0> <0.80,256,3.1> <1.50,256,4.9> <1.80,256,5.4> <0.20,256,9.8>

<0.30,256,4.6>

<1.40,256,2.9> <0.90,256,10.6> <0.60,256,12.5> <1.60,256,4.5> <0.50,256,6.9> <1.30,256,3.4> <1.00,256,5.2> <0.20,256,21.8> <1.20,256,13.7> <1.20,256,4.7>

<0.40,256,25.8>

<2.00,-256,6.8>

<0.20,256,5.5> <0.90,256,12.7> <1.00,256,7.1> <0.60,256,8.4> <0.30,256,13.4> <1.50,256,12.2> <1.30,256,8.1> <1.60,256,8.9> <1.00,256,8.6> <1.50,256,8.0> <1.50,256,13.4> <1.60,256,9.2> <1.30,256,8.2>

<1.00,256,17.0>

<0.50,256,12.0> <1.00,256,8.7>

<1.50,256,30.7>

<1.50,256,9.6> <0.80,256,11.2> <0.80,227,10.4>

<1.30,228,7.2>

<1.10,227,43.8> <1.50,256,4.1> <1.50,256,4.1> <1.50,256,4.1> <1.50,256,4.1>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-117
SLIDE 117

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.60,256,9.6> <1.00,-256,43.1>

<1.60,-256,55.1> <0.50,255,10.8> <1.60,252,63.2> <1.40,254,20.3> <1.10,256,3.4> <0.20,256,2.8> <0.60,256,3.0> <0.40,256,3.1> <2.00,256,4.0> <1.50,254,3.3> <0.90,256,2.8> <0.40,256,4.9> <0.90,256,3.0> <0.60,256,3.0> <1.00,256,6.7> <1.10,256,2.9> <0.70,256,3.1> <1.00,256,4.6> <1.80,256,3.4> <0.70,256,3.0> <0.70,256,3.1> <1.90,256,3.6> <0.80,256,3.0> <1.70,256,3.9> <0.70,256,4.4> <1.20,256,3.2> <1.70,256,4.3> <0.90,256,3.3> <1.10,256,2.5>

<1.00,256,6.3>

<0.60,256,3.4>

<0.50,256,5.3> <1.20,256,4.6> <0.40,256,4.0> <0.80,256,3.1> <1.50,256,4.9> <1.80,256,5.4> <1.50,256,4.1> <1.50,256,4.1> <1.50,256,4.1> <1.50,256,4.1> <0.20,256,9.8>

<0.30,256,4.6>

<1.40,256,2.9> <0.90,256,10.6> <0.60,256,12.5> <1.60,256,4.5> <0.50,256,6.9> <1.30,256,3.4> <1.00,256,5.2> <0.20,256,21.8> <1.20,256,13.7> <1.20,256,4.7>

<0.40,256,25.8>

<2.00,-256,6.8>

<0.20,256,5.5> <0.90,256,12.7> <1.00,256,7.1> <0.60,256,8.4> <0.30,256,13.4> <1.50,256,12.2> <1.30,256,8.1> <1.60,256,8.9> <1.00,256,8.6> <1.50,256,8.0> <1.50,256,13.4> <1.60,256,9.2> <1.30,256,8.2>

<1.00,256,17.0>

<0.50,256,12.0> <1.00,256,8.7>

<1.50,256,30.7>

<1.50,256,9.6> <0.80,256,11.2> <0.80,227,10.4>

<1.30,228,7.2>

<1.10,227,43.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-118
SLIDE 118

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.6> <0.90,-256,31.7>

<1.70,-256,48.8> <0.50,255,10.9> <1.60,252,57.7> <1.40,254,27.6> <1.10,256,3.2> <0.20,256,3.3> <0.60,256,3.0> <0.50,256,3.0> <2.00,256,3.3> <1.50,254,3.2> <0.90,256,2.9> <0.40,256,4.9> <0.90,256,4.2> <0.60,256,2.8> <1.00,256,5.9> <1.10,256,4.6> <0.70,256,3.9> <0.90,256,4.0> <1.70,256,3.3> <0.80,256,3.0> <0.50,256,4.0> <1.80,256,2.7> <0.60,256,3.0> <1.60,256,4.0> <0.80,256,4.9> <1.40,256,3.3> <1.80,256,5.8> <1.10,256,3.2> <0.60,256,2.9>

<1.20,256,5.7>

<0.80,256,3.5>

<0.50,256,3.6> <1.10,256,4.6> <0.50,256,3.8> <0.80,256,3.7> <1.70,256,6.5> <2.00,256,3.9> <1.50,256,4.2> <1.50,256,5.0> <1.50,256,9.4> <0.90,256,3.3> <0.10,256,12.1>

<0.10,256,3.8>

<1.40,256,2.9> <0.80,256,10.4> <0.70,256,7.0> <1.90,256,3.9> <1.10,256,7.8> <1.60,256,4.3> <0.50,256,6.0> <0.30,256,21.2> <1.40,256,7.3> <1.20,256,4.7>

<0.50,256,11.4>

<1.70,-256,6.5>

<0.40,256,5.6> <0.90,256,11.1> <1.00,256,15.6> <0.30,256,7.4> <0.40,256,11.8> <1.50,256,11.0> <1.40,256,7.7> <1.60,256,9.4> <0.50,256,9.3> <1.40,256,8.0> <1.40,256,25.3> <1.40,256,10.8> <1.40,256,8.1>

<0.70,256,15.5>

<0.60,256,12.7> <0.70,256,8.3>

<1.40,256,22.7>

<1.80,256,8.7> <0.50,256,11.2> <0.70,227,12.1>

<1.30,228,14.2>

<1.10,227,42.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-119
SLIDE 119

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.6> <0.90,-256,31.7>

<1.70,-256,48.8> <0.50,255,10.9> <1.60,252,57.7> <1.40,254,27.6> <1.10,256,3.2> <0.20,256,3.3> <0.60,256,3.0> <2.00,256,3.3> <1.50,254,3.2> <0.90,256,2.9> <0.40,256,4.9> <0.90,256,4.2> <0.60,256,2.8> <1.00,256,5.9> <1.10,256,4.6> <0.70,256,3.9> <0.90,256,4.0> <1.70,256,3.3> <0.80,256,3.0> <0.50,256,4.0> <1.80,256,2.7> <0.60,256,3.0> <1.60,256,4.0> <0.80,256,4.9> <1.40,256,3.3> <1.80,256,5.8> <1.10,256,3.2> <0.60,256,2.9>

<1.20,256,5.7>

<0.80,256,3.5>

<0.50,256,3.6> <1.10,256,4.6> <0.50,256,3.8> <0.80,256,3.7> <1.70,256,6.5> <2.00,256,3.9> <1.50,256,4.2> <1.50,256,5.0> <1.50,256,9.4> <0.90,256,3.3> <0.10,256,12.1>

<0.10,256,3.8>

<1.40,256,2.9> <0.80,256,10.4> <0.70,256,7.0> <1.90,256,3.9> <1.10,256,7.8> <1.60,256,4.3> <0.50,256,6.0> <0.30,256,21.2> <1.40,256,7.3> <1.20,256,4.7>

<0.50,256,11.4>

<1.70,-256,6.5>

<0.40,256,5.6> <0.90,256,11.1> <1.00,256,15.6> <0.30,256,7.4> <0.40,256,11.8> <1.50,256,11.0> <1.40,256,7.7> <1.60,256,9.4> <0.50,256,9.3> <1.40,256,8.0> <1.40,256,25.3> <1.40,256,10.8> <1.40,256,8.1>

<0.70,256,15.5>

<0.60,256,12.7> <0.70,256,8.3>

<1.40,256,22.7>

<1.80,256,8.7> <0.50,256,11.2> <0.70,227,12.1>

<1.30,228,14.2>

<1.10,227,42.8> <0.50,256,3.0> <0.50,256,3.0> <0.50,256,3.0> <0.50,256,3.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-120
SLIDE 120

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.6> <0.90,-256,31.7>

<1.70,-256,48.8> <0.50,255,10.9> <1.60,252,57.7> <1.40,254,27.6> <1.10,256,3.2> <0.20,256,3.3> <0.60,256,3.0> <0.50,256,3.0> <0.50,256,3.0> <0.50,256,3.0> <0.50,256,3.0> <2.00,256,3.3> <1.50,254,3.2> <0.90,256,2.9> <0.40,256,4.9> <0.90,256,4.2> <0.60,256,2.8> <1.00,256,5.9> <1.10,256,4.6> <0.70,256,3.9> <0.90,256,4.0> <1.70,256,3.3> <0.80,256,3.0> <0.50,256,4.0> <1.80,256,2.7> <0.60,256,3.0> <1.60,256,4.0> <0.80,256,4.9> <1.40,256,3.3> <1.80,256,5.8> <1.10,256,3.2> <0.60,256,2.9>

<1.20,256,5.7>

<0.80,256,3.5>

<0.50,256,3.6> <1.10,256,4.6> <0.50,256,3.8> <0.80,256,3.7> <1.70,256,6.5> <2.00,256,3.9> <1.50,256,4.2> <1.50,256,5.0> <1.50,256,9.4> <0.90,256,3.3> <0.10,256,12.1>

<0.10,256,3.8>

<1.40,256,2.9> <0.80,256,10.4> <0.70,256,7.0> <1.90,256,3.9> <1.10,256,7.8> <1.60,256,4.3> <0.50,256,6.0> <0.30,256,21.2> <1.40,256,7.3> <1.20,256,4.7>

<0.50,256,11.4>

<1.70,-256,6.5>

<0.40,256,5.6> <0.90,256,11.1> <1.00,256,15.6> <0.30,256,7.4> <0.40,256,11.8> <1.50,256,11.0> <1.40,256,7.7> <1.60,256,9.4> <0.50,256,9.3> <1.40,256,8.0> <1.40,256,25.3> <1.40,256,10.8> <1.40,256,8.1>

<0.70,256,15.5>

<0.60,256,12.7> <0.70,256,8.3>

<1.40,256,22.7>

<1.80,256,8.7> <0.50,256,11.2> <0.70,227,12.1>

<1.30,228,14.2>

<1.10,227,42.8>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-121
SLIDE 121

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.1> <0.90,-256,26.0>

<1.70,-256,41.5> <0.50,255,3.1> <1.60,252,8.9> <1.40,254,22.0> <1.10,256,2.9> <0.10,256,4.0> <0.60,256,3.8> <0.50,256,3.1> <0.50,256,6.1> <0.50,256,3.0> <0.30,256,3.0> <2.00,256,4.1> <1.60,254,3.0> <0.90,256,3.0> <0.40,256,2.9> <0.90,256,2.6> <0.60,256,2.9> <1.10,256,3.4> <1.10,256,3.9> <0.70,256,2.9> <0.90,256,4.0> <1.50,256,3.4> <0.70,256,2.9> <0.50,256,5.4> <1.90,256,4.5> <0.70,256,3.1> <1.60,256,3.0> <0.80,256,4.2> <1.30,256,3.4> <1.40,256,5.7> <1.30,256,4.2> <0.50,256,3.0>

<1.20,256,5.1>

<0.70,256,4.2>

<0.50,256,3.5> <1.00,256,3.6> <0.60,256,5.2> <0.30,256,3.9> <1.60,256,6.4> <1.70,256,3.7> <1.70,256,3.3> <1.40,256,4.1> <1.70,256,10.4> <1.00,256,3.0> <0.10,256,12.1>

<0.10,256,4.9>

<1.70,256,2.9> <0.50,256,11.3> <1.40,256,8.1> <1.50,256,3.8> <0.90,256,9.3> <1.50,256,4.5> <0.80,256,5.8> <0.10,256,16.4> <1.30,256,7.8> <1.10,256,3.8>

<1.00,256,24.2>

<1.70,-256,4.8>

<0.10,256,6.5> <0.80,256,11.5> <0.90,256,14.6> <0.40,256,6.5> <0.20,256,13.5> <1.50,256,11.3> <1.50,256,7.4> <1.40,256,7.7> <0.40,256,11.9> <1.30,256,7.6> <1.50,256,31.7> <1.40,256,17.1> <1.50,256,7.1>

<0.70,256,15.5>

<0.50,256,6.2> <0.70,256,8.4>

<1.40,256,22.6>

<1.80,256,10.0> <0.50,256,9.8> <0.70,227,62.7>

<1.30,228,18.8>

<1.10,227,29.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-122
SLIDE 122

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.1> <0.90,-256,26.0>

<1.70,-256,41.5> <0.50,255,3.1> <1.60,252,8.9> <1.40,254,22.0> <0.10,256,4.0> <0.60,256,3.8> <0.50,256,3.1> <0.50,256,6.1> <0.50,256,3.0> <0.30,256,3.0> <2.00,256,4.1> <1.60,254,3.0> <0.90,256,3.0> <0.40,256,2.9> <0.90,256,2.6> <0.60,256,2.9> <1.10,256,3.4> <1.10,256,3.9> <0.70,256,2.9> <0.90,256,4.0> <1.50,256,3.4> <0.70,256,2.9> <0.50,256,5.4> <1.90,256,4.5> <0.70,256,3.1> <1.60,256,3.0> <0.80,256,4.2> <1.30,256,3.4> <1.40,256,5.7> <1.30,256,4.2> <0.50,256,3.0>

<1.20,256,5.1>

<0.70,256,4.2>

<0.50,256,3.5> <1.00,256,3.6> <0.60,256,5.2> <0.30,256,3.9> <1.60,256,6.4> <1.70,256,3.7> <1.70,256,3.3> <1.40,256,4.1> <1.70,256,10.4> <1.00,256,3.0> <0.10,256,12.1>

<0.10,256,4.9>

<1.70,256,2.9> <0.50,256,11.3> <1.40,256,8.1> <1.50,256,3.8> <0.90,256,9.3> <1.50,256,4.5> <0.80,256,5.8> <0.10,256,16.4> <1.30,256,7.8> <1.10,256,3.8>

<1.00,256,24.2>

<1.70,-256,4.8>

<0.10,256,6.5> <0.80,256,11.5> <0.90,256,14.6> <0.40,256,6.5> <0.20,256,13.5> <1.50,256,11.3> <1.50,256,7.4> <1.40,256,7.7> <0.40,256,11.9> <1.30,256,7.6> <1.50,256,31.7> <1.40,256,17.1> <1.50,256,7.1>

<0.70,256,15.5>

<0.50,256,6.2> <0.70,256,8.4>

<1.40,256,22.6>

<1.80,256,10.0> <0.50,256,9.8> <0.70,227,62.7>

<1.30,228,18.8>

<1.10,227,29.2> <1.10,256,2.9> <1.10,256,2.9> <1.10,256,2.9> <1.10,256,2.9>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-123
SLIDE 123

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.90,256,9.1> <0.90,-256,26.0>

<1.70,-256,41.5> <0.50,255,3.1> <1.60,252,8.9> <1.40,254,22.0> <1.10,256,2.9> <1.10,256,2.9> <1.10,256,2.9> <1.10,256,2.9> <0.10,256,4.0> <0.60,256,3.8> <0.50,256,3.1> <0.50,256,6.1> <0.50,256,3.0> <0.30,256,3.0> <2.00,256,4.1> <1.60,254,3.0> <0.90,256,3.0> <0.40,256,2.9> <0.90,256,2.6> <0.60,256,2.9> <1.10,256,3.4> <1.10,256,3.9> <0.70,256,2.9> <0.90,256,4.0> <1.50,256,3.4> <0.70,256,2.9> <0.50,256,5.4> <1.90,256,4.5> <0.70,256,3.1> <1.60,256,3.0> <0.80,256,4.2> <1.30,256,3.4> <1.40,256,5.7> <1.30,256,4.2> <0.50,256,3.0>

<1.20,256,5.1>

<0.70,256,4.2>

<0.50,256,3.5> <1.00,256,3.6> <0.60,256,5.2> <0.30,256,3.9> <1.60,256,6.4> <1.70,256,3.7> <1.70,256,3.3> <1.40,256,4.1> <1.70,256,10.4> <1.00,256,3.0> <0.10,256,12.1>

<0.10,256,4.9>

<1.70,256,2.9> <0.50,256,11.3> <1.40,256,8.1> <1.50,256,3.8> <0.90,256,9.3> <1.50,256,4.5> <0.80,256,5.8> <0.10,256,16.4> <1.30,256,7.8> <1.10,256,3.8>

<1.00,256,24.2>

<1.70,-256,4.8>

<0.10,256,6.5> <0.80,256,11.5> <0.90,256,14.6> <0.40,256,6.5> <0.20,256,13.5> <1.50,256,11.3> <1.50,256,7.4> <1.40,256,7.7> <0.40,256,11.9> <1.30,256,7.6> <1.50,256,31.7> <1.40,256,17.1> <1.50,256,7.1>

<0.70,256,15.5>

<0.50,256,6.2> <0.70,256,8.4>

<1.40,256,22.6>

<1.80,256,10.0> <0.50,256,9.8> <0.70,227,62.7>

<1.30,228,18.8>

<1.10,227,29.2>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-124
SLIDE 124

Introduction The problem Remy Evaluation Discussion

RemyCC

s_ewma r_ewma

<1.60,256,10.9> <0.90,-256,32.4>

<1.50,-256,41.5> <0.40,255,3.0> <1.60,252,5.7> <1.40,254,2.7> <1.10,256,3.0> <1.10,256,2.0> <1.00,256,4.2> <1.10,256,2.9> <0.10,256,3.3> <0.60,256,5.4> <0.50,256,3.7> <0.50,256,7.4> <0.50,256,4.5> <0.40,256,2.8> <2.00,256,3.7> <1.60,254,2.6> <0.90,256,3.3> <0.50,256,2.8> <0.90,256,4.0> <0.60,256,2.6> <1.10,256,4.7> <0.90,256,3.2> <0.70,256,3.7> <0.80,256,4.2> <1.40,256,3.4> <0.50,256,2.9> <0.70,256,4.5> <1.90,256,3.6> <0.90,256,3.0> <1.90,256,3.8> <0.70,256,3.7> <1.40,256,3.5> <1.20,256,6.4> <1.20,256,3.4> <0.70,256,2.9>

<1.40,256,5.5>

<0.80,256,3.5>

<0.60,256,3.4> <1.10,256,3.7> <0.50,256,3.9> <0.40,256,4.0> <1.30,256,6.5> <1.60,256,3.6> <1.80,256,4.9> <1.40,256,3.5> <1.60,256,4.7> <1.20,256,2.3> <0.20,256,13.1>

<0.10,256,4.0>

<1.80,256,3.1> <0.50,256,12.9> <1.60,256,8.9> <1.80,256,3.9> <0.30,256,8.6> <1.70,256,6.4> <0.70,256,5.4> <0.10,256,15.6> <0.80,256,6.1> <1.00,256,3.6>

<1.00,256,26.0>

<1.60,-256,6.3>

<0.10,256,8.3> <0.50,256,19.4> <0.70,256,14.7> <0.10,256,5.8> <0.10,256,15.3> <1.70,256,12.2> <1.60,256,5.6> <1.20,256,7.6> <0.60,256,12.7> <1.20,256,9.1> <1.20,256,17.8> <1.30,256,9.9> <1.30,256,7.9>

<0.80,256,13.6>

<0.90,256,7.9> <0.90,256,9.6>

<1.40,256,35.4>

<1.80,256,9.2> <0.40,256,10.7> <0.70,227,5.8>

<1.30,228,19.6>

<1.10,227,25.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-125
SLIDE 125

Introduction The problem Remy Evaluation Discussion

RemyCC

s_ewma r_ewma

<1.60,256,10.9> <0.90,-256,32.4>

<1.50,-256,41.5> <0.40,255,3.0> <1.60,252,5.7> <1.40,254,2.7> <1.10,256,3.0> <1.10,256,2.0> <1.00,256,4.2> <1.10,256,2.9> <0.10,256,3.3> <0.60,256,5.4> <0.50,256,3.7> <0.50,256,7.4> <0.50,256,4.5> <0.40,256,2.8> <2.00,256,3.7> <1.60,254,2.6> <0.90,256,3.3> <0.50,256,2.8> <0.90,256,4.0> <0.60,256,2.6> <1.10,256,4.7> <0.90,256,3.2> <0.70,256,3.7> <0.80,256,4.2> <1.40,256,3.4> <0.50,256,2.9> <0.70,256,4.5> <1.90,256,3.6> <1.90,256,3.8> <0.70,256,3.7> <1.40,256,3.5> <1.20,256,6.4> <1.20,256,3.4> <0.70,256,2.9>

<1.40,256,5.5>

<0.80,256,3.5>

<0.60,256,3.4> <1.10,256,3.7> <0.50,256,3.9> <0.40,256,4.0> <1.30,256,6.5> <1.60,256,3.6> <1.80,256,4.9> <1.40,256,3.5> <1.60,256,4.7> <1.20,256,2.3> <0.20,256,13.1>

<0.10,256,4.0>

<1.80,256,3.1> <0.50,256,12.9> <1.60,256,8.9> <1.80,256,3.9> <0.30,256,8.6> <1.70,256,6.4> <0.70,256,5.4> <0.10,256,15.6> <0.80,256,6.1> <1.00,256,3.6>

<1.00,256,26.0>

<1.60,-256,6.3>

<0.10,256,8.3> <0.50,256,19.4> <0.70,256,14.7> <0.10,256,5.8> <0.10,256,15.3> <1.70,256,12.2> <1.60,256,5.6> <1.20,256,7.6> <0.60,256,12.7> <1.20,256,9.1> <1.20,256,17.8> <1.30,256,9.9> <1.30,256,7.9>

<0.80,256,13.6>

<0.90,256,7.9> <0.90,256,9.6>

<1.40,256,35.4>

<1.80,256,9.2> <0.40,256,10.7> <0.70,227,5.8>

<1.30,228,19.6>

<1.10,227,25.3> <0.90,256,3.0> <0.90,256,3.0> <0.90,256,3.0> <0.90,256,3.0>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-126
SLIDE 126

Introduction The problem Remy Evaluation Discussion

RemyCC .

s_ewma r_ewma

<1.60,256,10.9> <0.90,-256,32.4>

<1.50,-256,41.5> <0.40,255,3.0> <1.60,252,5.7> <1.40,254,2.7> <1.10,256,3.0> <1.10,256,2.0> <1.00,256,4.2> <1.10,256,2.9> <0.10,256,3.3> <0.60,256,5.4> <0.50,256,3.7> <0.50,256,7.4> <0.50,256,4.5> <0.40,256,2.8> <2.00,256,3.7> <1.60,254,2.6> <0.90,256,3.3> <0.50,256,2.8> <0.90,256,4.0> <0.60,256,2.6> <1.10,256,4.7> <0.90,256,3.2> <0.70,256,3.7> <0.80,256,4.2> <1.40,256,3.4> <0.50,256,2.9> <0.70,256,4.5> <1.90,256,3.6> <0.90,256,3.0> <0.90,256,3.0> <0.90,256,3.0> <0.90,256,3.0> <1.90,256,3.8> <0.70,256,3.7> <1.40,256,3.5> <1.20,256,6.4> <1.20,256,3.4> <0.70,256,2.9>

<1.40,256,5.5>

<0.80,256,3.5>

<0.60,256,3.4> <1.10,256,3.7> <0.50,256,3.9> <0.40,256,4.0> <1.30,256,6.5> <1.60,256,3.6> <1.80,256,4.9> <1.40,256,3.5> <1.60,256,4.7> <1.20,256,2.3> <0.20,256,13.1>

<0.10,256,4.0>

<1.80,256,3.1> <0.50,256,12.9> <1.60,256,8.9> <1.80,256,3.9> <0.30,256,8.6> <1.70,256,6.4> <0.70,256,5.4> <0.10,256,15.6> <0.80,256,6.1> <1.00,256,3.6>

<1.00,256,26.0>

<1.60,-256,6.3>

<0.10,256,8.3> <0.50,256,19.4> <0.70,256,14.7> <0.10,256,5.8> <0.10,256,15.3> <1.70,256,12.2> <1.60,256,5.6> <1.20,256,7.6> <0.60,256,12.7> <1.20,256,9.1> <1.20,256,17.8> <1.30,256,9.9> <1.30,256,7.9>

<0.80,256,13.6>

<0.90,256,7.9> <0.90,256,9.6>

<1.40,256,35.4>

<1.80,256,9.2> <0.40,256,10.7> <0.70,227,5.8>

<1.30,228,19.6>

<1.10,227,25.3>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-127
SLIDE 127

Introduction The problem Remy Evaluation Discussion

RemyCC

s_ewma r_ewma

<1.70,256,10.4> <0.70,-256,44.4>

<1.10,-256,5.0> <0.10,255,2.7> <1.60,252,3.0> <1.40,254,1.8> <1.30,256,3.4> <1.20,256,4.5> <1.00,256,10.8> <1.00,256,3.2> <0.10,256,3.2> <0.50,256,3.7> <0.50,256,3.8> <0.40,256,5.7> <0.70,256,2.9> <0.20,256,2.3> <2.00,256,3.5> <1.80,254,2.6> <0.90,256,3.9> <0.50,256,2.0> <0.90,256,4.0> <0.50,256,2.9> <1.10,256,4.6> <1.00,256,3.8> <0.80,256,3.7> <0.80,256,4.9> <1.40,256,4.2> <0.40,256,2.7> <0.50,256,4.5> <2.00,256,2.7> <1.20,256,2.7> <1.00,256,3.1> <0.90,256,2.5> <0.40,256,3.1> <1.90,256,3.0> <1.00,256,3.9> <1.70,256,4.1> <1.60,256,4.8> <1.10,256,4.3> <1.20,256,3.0>

<1.90,256,5.6>

<0.70,256,3.7>

<0.50,256,3.3> <1.20,256,5.1> <0.20,256,5.5> <0.30,256,3.9> <1.60,256,8.3> <1.50,256,2.5> <1.60,256,4.7> <1.60,256,3.4> <1.40,256,5.5> <0.70,256,3.0> <0.50,256,12.9>

<0.20,256,4.9>

<1.70,256,3.2> <0.40,256,7.4> <1.30,256,3.6> <1.90,256,3.0> <0.10,256,8.6> <1.60,256,5.4> <0.50,256,4.5> <0.10,256,20.1> <0.60,256,11.8> <1.10,256,5.1>

<1.00,256,12.3>

<1.80,-256,5.4>

<0.40,256,6.7> <0.50,256,20.2> <1.00,256,9.3> <0.30,256,4.1> <0.30,256,22.6> <1.60,256,10.6> <1.50,256,7.3> <1.40,256,6.7> <0.60,256,14.2> <1.70,256,8.3> <0.80,256,23.4> <1.70,256,9.9> <1.20,256,7.0>

<1.20,256,13.1>

<0.70,256,7.9> <1.00,256,9.9>

<1.40,256,43.4>

<1.60,256,10.9> <0.10,256,9.9> <0.70,227,56.1>

<1.30,228,27.0>

<1.10,227,19.6>

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-128
SLIDE 128

Introduction The problem Remy Evaluation Discussion

Evaluation in ns-2

◮ End-to-end comparators: NewReno, Cubic, Compound, Vegas ◮ In-net comparators: Cubic-over-sfqCoDel, XCP ◮ Simulation setup published for replication

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-129
SLIDE 129

Introduction The problem Remy Evaluation Discussion

Scenario 1: fixed-rate network, homogenous senders

...

Sender Sender 2 Sender n Queue

...

Receiver Receiver 2 Receiver n Round-trip time Link

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-130
SLIDE 130

Introduction The problem Remy Evaluation Discussion

Scenario 1: details

Quantity Simulation parameter Remy assumptions Link speed 15 Mbps Uniform(10, 20) Mbps RTT 150 ms Uniform(100, 200) ms n 8 Uniform(1, 16) “On” process exp[µ = 100] kB exp[µ = 5] s “Off” process exp

  • µ = 1

2

  • s

exp[µ = 5] s Remy objective:

  • i

log

  • throughputi
  • delayi

δ

  • δ ∈

1 10, 1, 10

  • Keith Winstein and Hari Balakrishnan

MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-131
SLIDE 131

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-132
SLIDE 132

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-133
SLIDE 133

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-134
SLIDE 134

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-135
SLIDE 135

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-136
SLIDE 136

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better

Delay unfairness Throughput unfairness Median

  • utcome

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-137
SLIDE 137

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-138
SLIDE 138

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-139
SLIDE 139

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno Vegas Cubic Compound

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-140
SLIDE 140

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno Vegas Cubic Compound XCP Cubic/sfqCoDel

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-141
SLIDE 141

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno Vegas Cubic Compound XCP Cubic/sfqCoDel RemyCC =1 RemyCC δ=10 δ δ=0.1 RemyCC

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-142
SLIDE 142

Introduction The problem Remy Evaluation Discussion

Scenario 1: throughput-delay plot

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1 2 4 8 16 32 Throughput (Mbps) Queueing delay (ms)

Better NewReno Vegas Cubic Compound XCP Cubic/sfqCoDel RemyCC =1 RemyCC δ=10 δ δ=0.1 RemyCC

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-143
SLIDE 143

Introduction The problem Remy Evaluation Discussion

Scenario 2: Verizon LTE, n = 8

0.8 1 1.2 1.4 1.6 1.8 2 16 32 64 Throughput (Mbps) Queueing delay (ms)

Vegas Remy δ=0.1 Remy δ=1 Remy δ=10 Cubic Compound NewReno XCP Cubic/sfqCoDel

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-144
SLIDE 144

Introduction The problem Remy Evaluation Discussion

The effect of prior knowledge

  • 6
  • 5
  • 4
  • 3
  • 2
  • 1

4.74 15 47.4 log(normalized throughput) - log(delay) link speed (megabits/sec) RemyCC 10x

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-145
SLIDE 145

Introduction The problem Remy Evaluation Discussion

The effect of prior knowledge

  • 6
  • 5
  • 4
  • 3
  • 2
  • 1

4.74 15 47.4 log(normalized throughput) - log(delay) link speed (megabits/sec) RemyCC 10x RemyCC exact

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-146
SLIDE 146

Introduction The problem Remy Evaluation Discussion

The effect of prior knowledge

  • 6
  • 5
  • 4
  • 3
  • 2
  • 1

4.74 15 47.4 log(normalized throughput) - log(delay) link speed (megabits/sec) RemyCC 10x RemyCC exact Cubic-

  • ver-

sfqCoDel

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-147
SLIDE 147

Introduction The problem Remy Evaluation Discussion

Limitations and unknowns

◮ Tested only in simulation so far ◮ How to characterize robustness to the unforeseen? ◮ Can we make a RemyCC for a 10,000x range of throughputs? ◮ Agreeing on assumptions and goal may not be easy ◮ End-to-end = hard to handle an overaggressive competitor ◮ Not proposing Internet-scale deployment any time soon

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control

slide-148
SLIDE 148

Introduction The problem Remy Evaluation Discussion

Conclusions

◮ Traditionally: simple rules, complex behavior ◮ With Remy: complex rules, consistent behavior ◮ Computer-designed > human-designed ◮ End-to-end > in-network ◮ The network evolves. Transport should let it!

http://web.mit.edu/remy {keithw, hari}@mit.edu

Keith Winstein and Hari Balakrishnan MIT Computer Science and Artificial Intelligence Laboratory TCP ex Machina: Computer-Generated Congestion Control