an experimental study of the learnability of congestion...
TRANSCRIPT
![Page 1: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/1.jpg)
An experimental study of the learnability ofcongestion control
Anirudh Sivaraman, Keith Winstein, Pratiksha Thaker,Hari Balakrishnan
MIT CSAIL
http://web.mit.edu/remy/learnability
August 31, 2014
1 / 17
![Page 2: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/2.jpg)
This talk
I How easy is it to learn a network protocol toachieve a desired goal, despite a mismatched setof assumptions?
I cf. Learning: “Knowledge acquisition withoutexplicit programming” (Valiant 1984)
2 / 17
![Page 3: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/3.jpg)
This talk
I How easy is it to learn a network protocol toachieve a desired goal, despite a mismatched setof assumptions?
I cf. Learning: “Knowledge acquisition withoutexplicit programming” (Valiant 1984)
2 / 17
![Page 4: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/4.jpg)
Preview of key results
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
3 / 17
![Page 5: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/5.jpg)
Preview of key results
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
3 / 17
![Page 6: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/6.jpg)
Preview of key results
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
3 / 17
![Page 7: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/7.jpg)
Preview of key results
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
3 / 17
![Page 8: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/8.jpg)
Preview of key results
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
3 / 17
![Page 9: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/9.jpg)
Experimental method
4 / 17
![Page 10: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/10.jpg)
Experimental method
4 / 17
![Page 11: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/11.jpg)
Experimental method
< Mbps, ms>
4 / 17
![Page 12: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/12.jpg)
Experimental method
< Mbps, ms>
4 / 17
![Page 13: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/13.jpg)
Experimental method
< Mbps, ms>
4 / 17
![Page 14: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/14.jpg)
Experimental method
< Mbps, ms>
4 / 17
![Page 15: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/15.jpg)
Experimental method
< Mbps, ms>
4 / 17
![Page 16: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/16.jpg)
Experimental method
< Mbps, ms>
Training Networks
5 / 17
![Page 17: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/17.jpg)
Experimental method
< Mbps, ms>
Training Networks
Objective Function:- log (tpt/delay)- Avg. Flow Completion time
Learner
5 / 17
![Page 18: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/18.jpg)
Experimental method
< Mbps, ms>
Training Networks
Objective Function:- log (tpt/delay)- Avg. Flow Completion time
LearnerCongestionControlAlgorithm
5 / 17
![Page 19: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/19.jpg)
Experimental method
< Mbps, ms>
Training Networks
Objective Function:- log (tpt/delay)- Avg. Flow Completion time
Remy(SIGCOMM 13)
RemyCC
5 / 17
![Page 20: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/20.jpg)
Experimental method
< Mbps, ms>
Training Networks
< Mbps, ms>
Test withinns-2
Testing Networks
Objective Function:- log (tpt/delay)- Avg. Flow Completion time
Remy(SIGCOMM 13)
RemyCC
5 / 17
![Page 21: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/21.jpg)
Remy compared with an ideal protocol
0.5
1
2
4
8
16
32
0100200300400500
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
6 / 17
![Page 22: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/22.jpg)
Remy compared with an ideal protocol
0.5
1
2
4
8
16
32
0100200300400500
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Ideal
6 / 17
![Page 23: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/23.jpg)
Remy compared with an ideal protocol
0.5
1
2
4
8
16
32
0100200300400500
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Ideal
RemyCC
6 / 17
![Page 24: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/24.jpg)
Remy compared with an ideal protocol
0.5
1
2
4
8
16
32
0100200300400500
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Ideal
RemyCC
Cubic Cubic/sfqCoDel
6 / 17
![Page 25: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/25.jpg)
Learning network protocols despite mismatchedassumptions
I Is there a tradeoff between operating range andgenerality in link rates?
I Is there a tradeoff between performance andoperating range in link rates?
7 / 17
![Page 26: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/26.jpg)
Learning network protocols despite mismatchedassumptions
I Is there a tradeoff between operating range andgenerality in link rates?
I Is there a tradeoff between performance andoperating range in link rates?
7 / 17
![Page 27: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/27.jpg)
Learning network protocols despite mismatchedassumptions
I Is there a tradeoff between operating range andgenerality in link rates?
I Is there a tradeoff between performance andoperating range in link rates?
7 / 17
![Page 28: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/28.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
8 / 17
![Page 29: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/29.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
8 / 17
![Page 30: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/30.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
2x range
8 / 17
![Page 31: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/31.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
2x range10x range
8 / 17
![Page 32: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/32.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
2x range10x range
100x range
8 / 17
![Page 33: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/33.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
2x range10x range
100x range1000x range
8 / 17
![Page 34: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/34.jpg)
Performance and link-rate operating range
-1.5
-1
-0.5
0
1 10 100 1000Link rate (Mbps)
Objective Function(Normalized)
Ideal
Cubic
Cubic-over-sfqCoDel
2x range10x range
100x range1000x range
8 / 17
![Page 35: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/35.jpg)
Performance and link-rate operating range
I Very clear generality vs. operating range tradeoff
I Only weak evidence of a performance vs.operating range tradeoff
I Possible to design a forwards-comptabibleprotocol handling a wide range in link rates
9 / 17
![Page 36: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/36.jpg)
Performance and link-rate operating range
I Very clear generality vs. operating range tradeoff
I Only weak evidence of a performance vs.operating range tradeoff
I Possible to design a forwards-comptabibleprotocol handling a wide range in link rates
9 / 17
![Page 37: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/37.jpg)
Performance and link-rate operating range
I Very clear generality vs. operating range tradeoff
I Only weak evidence of a performance vs.operating range tradeoff
I Possible to design a forwards-comptabibleprotocol handling a wide range in link rates
9 / 17
![Page 38: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/38.jpg)
Performance and link-rate operating range
I Very clear generality vs. operating range tradeoff
I Only weak evidence of a performance vs.operating range tradeoff
I Possible to design a forwards-comptabibleprotocol handling a wide range in link rates
9 / 17
![Page 39: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/39.jpg)
Learning network protocols despite mismatchedassumptions
Can we learn a protocol that performs well bothwhen there are few senders and when there aremany senders?
10 / 17
![Page 40: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/40.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
11 / 17
![Page 41: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/41.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
11 / 17
![Page 42: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/42.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
11 / 17
![Page 43: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/43.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
11 / 17
![Page 44: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/44.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
11 / 17
![Page 45: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/45.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
11 / 17
![Page 46: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/46.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
11 / 17
![Page 47: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/47.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
11 / 17
![Page 48: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/48.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
11 / 17
![Page 49: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/49.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
1 - 100
11 / 17
![Page 50: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/50.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
1 - 100
11 / 17
![Page 51: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/51.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
1 - 100
Cubic
11 / 17
![Page 52: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/52.jpg)
Imperfections in the number of senders
0 20 40 60 80 100
Number of senders
−1.4
−1.2
−1.0
−0.8
−0.6
−0.4
−0.2
0.0
Nor
mal
ized
obje
ctiv
efu
nctio
n
Ideal
1-2
1- 10
1 - 50
1 - 100
Cubic
Cubic-over-sfqCoDel
11 / 17
![Page 53: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/53.jpg)
Imperfections in the number of senders
Tradeoff between performance with few senders andperformance with many senders
11 / 17
![Page 54: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/54.jpg)
Learning network protocols despite mismatchedassumptions
What are the costs and benefits of learning a newprotocol that shares fairly with a legacy sender?
12 / 17
![Page 55: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/55.jpg)
Imperfect assumptions about the nature of other senders
I TCP-Aware RemyCC: Contends with:I TCP-Aware RemyCC half the timeI TCP NewReno half the time.
I TCP-Naive RemyCC: Contends with:I TCP-Naive RemyCC all the time
13 / 17
![Page 56: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/56.jpg)
Imperfect assumptions about the nature of other senders
I TCP-Aware RemyCC: Contends with:I TCP-Aware RemyCC half the timeI TCP NewReno half the time.
I TCP-Naive RemyCC: Contends with:I TCP-Naive RemyCC all the time
13 / 17
![Page 57: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/57.jpg)
RemyCC competing against itself
3
4
5
6
7
163264128
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Better
NewRenoRemyCC
[TCP-naive]
14 / 17
![Page 58: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/58.jpg)
RemyCC competing against itself
3
4
5
6
7
163264128
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Better
NewRenoRemyCC
[TCP-naive]
Cost of TCP-awareness
14 / 17
![Page 59: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/59.jpg)
RemyCC competing against itself
3
4
5
6
7
163264128
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Better
NewRenoRemyCC
[TCP-naive]
Cost of TCP-awareness
RemyCC[TCP-aware]
14 / 17
![Page 60: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/60.jpg)
RemyCC competing against TCP NewReno
4
5
6
7
6496128
Queueing delay (ms)
Better
3
Thr
ough
put (
Mbp
s)
NewReno
RemyCC[TCP-naive]
15 / 17
![Page 61: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/61.jpg)
RemyCC competing against TCP NewReno
4
5
6
7
6496128
Queueing delay (ms)
Better
3
Thr
ough
put (
Mbp
s)
NewReno
RemyCC[TCP-naive]
Benefit of TCP-awareness
Effect ofTCP-awareadversary
15 / 17
![Page 62: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/62.jpg)
RemyCC competing against TCP NewReno
4
5
6
7
6496128
Queueing delay (ms)
Better
3
Thr
ough
put (
Mbp
s)
NewRenoRemyCC
[TCP-aware]
NewReno
RemyCC[TCP-naive]
Benefit of TCP-awareness
Effect ofTCP-awareadversary
15 / 17
![Page 63: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/63.jpg)
RemyCC competing against TCP NewReno
TCP awareness benefits you when needed, costs ifyou don’t
15 / 17
![Page 64: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/64.jpg)
Caveats
I Remy as a proxy for an optimal learner
I Results may change with better learners
I Negative results may no longer hold
16 / 17
![Page 65: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/65.jpg)
Caveats
I Remy as a proxy for an optimal learner
I Results may change with better learners
I Negative results may no longer hold
16 / 17
![Page 66: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/66.jpg)
Caveats
I Remy as a proxy for an optimal learner
I Results may change with better learners
I Negative results may no longer hold
16 / 17
![Page 67: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/67.jpg)
Caveats
I Remy as a proxy for an optimal learner
I Results may change with better learners
I Negative results may no longer hold
16 / 17
![Page 68: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/68.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 69: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/69.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 70: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/70.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 71: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/71.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 72: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/72.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecks
I Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 73: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/73.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 74: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/74.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transport
I user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 75: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/75.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 76: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/76.jpg)
The learnability of congestion control
I Can tolerate mismatched link-rate assumptions
I Need precision about the number of senders
I TCP compatibility is a double-edged sword
I Can tolerate mismatch in the # of bottlenecksI Ongoing work in using findings:
I improve Google’s datacenter transportI user-space implementation of RemyCC
I http://web.mit.edu/remy/learnability
17 / 17
![Page 77: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/77.jpg)
Backup slides
17 / 17
![Page 78: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/78.jpg)
The Remy protocol synthesis procedure
I Protocol: range-based rule table from state to action
I State: Congestion signals tracked by the senderI s ewma : EWMA over packet inter-transmit timesI r ewma : EWMA over ACK inter-arrival timesI rtt ratio: Ratio of RTT to minimum RTTI slow r ewma: Slower version of s ewma
I Action: modify window, transmission rateI Multiplier m to current windowI Increment c to current windowI Minimum inter-transmit time.
17 / 17
![Page 79: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/79.jpg)
The Remy protocol synthesis procedure
I Protocol: range-based rule table from state to actionI State: Congestion signals tracked by the sender
I s ewma : EWMA over packet inter-transmit timesI r ewma : EWMA over ACK inter-arrival timesI rtt ratio: Ratio of RTT to minimum RTTI slow r ewma: Slower version of s ewma
I Action: modify window, transmission rateI Multiplier m to current windowI Increment c to current windowI Minimum inter-transmit time.
17 / 17
![Page 80: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/80.jpg)
The Remy protocol synthesis procedure
I Protocol: range-based rule table from state to actionI State: Congestion signals tracked by the sender
I s ewma : EWMA over packet inter-transmit timesI r ewma : EWMA over ACK inter-arrival timesI rtt ratio: Ratio of RTT to minimum RTTI slow r ewma: Slower version of s ewma
I Action: modify window, transmission rateI Multiplier m to current windowI Increment c to current windowI Minimum inter-transmit time.
17 / 17
![Page 81: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/81.jpg)
The Remy protocol synthesis procedure
1. Start with one rule: one action for all states
2. Optimize each action to maximize objective
3. Find most used rule
4. Median split that rule based on state usage
5. Repeat 2, 3, and 4 till you converge
17 / 17
![Page 82: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/82.jpg)
One action for all states. Find the best value.
s_ewma
r_ewma
<?,?,?>
17 / 17
![Page 83: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/83.jpg)
The best (single) action. Now split it on median.
s_ewma
r_ewma
<0.90,4,3.3>
17 / 17
![Page 84: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/84.jpg)
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>
17 / 17
![Page 85: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/85.jpg)
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>
17 / 17
![Page 86: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/86.jpg)
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>
17 / 17
![Page 87: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/87.jpg)
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>
17 / 17
![Page 88: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/88.jpg)
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>
17 / 17
![Page 89: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/89.jpg)
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>
17 / 17
![Page 90: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/90.jpg)
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>
17 / 17
![Page 91: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/91.jpg)
Can applications with different objectives coexist?
I Tpt. Sender: A throughput-intensive sender
log(throughput)− 0.1 ∗ log(delay) (1)
I Lat. Sender: A latency-sensitive sender
log(throughput)− 10.0 ∗ log(delay) (2)
I Running over a FIFO queue
17 / 17
![Page 92: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/92.jpg)
Training for diversity has a cost ...
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
17 / 17
![Page 93: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/93.jpg)
Training for diversity has a cost ...
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Tpt. Sender[naive]
Lat. Sender[naive]
17 / 17
![Page 94: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/94.jpg)
Training for diversity has a cost ...
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Tpt. Sender[naive]
Tpt. Sender[coevolved]
Lat. Sender[naive]
Lat. Sender[coevolved]
Cost of Coexistence
17 / 17
![Page 95: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/95.jpg)
but, benefits the docile sender
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
17 / 17
![Page 96: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/96.jpg)
but, benefits the docile sender
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Tpt. Sender[naive]
Lat. Sender[naive]
17 / 17
![Page 97: An experimental study of the learnability of congestion ...web.mit.edu/Remy/Learnability-slides.pdfAn experimental study of the learnability of congestion control Anirudh Sivaraman,](https://reader035.vdocuments.us/reader035/viewer/2022062602/5ec0f9f7dabe131a5c38e26f/html5/thumbnails/97.jpg)
but, benefits the docile sender
1
2
5
11
16
124816326412825651210242048
Thr
ough
put (
Mbp
s)
Queueing delay (ms)
Tpt. Sender[naive]
Tpt. Sender[coevolved]
Lat. Sender[naive]
Lat. Sender[coevolved]
Benefit of coevolution
Effect ofplaying nice
17 / 17