Download - TCP Pacing in Data Center Networks
![Page 1: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/1.jpg)
TCP Pacing in Data Center Networks
1
Monia Ghobadi, Yashar Ganjali
Department of Computer Science, University of Toronto{monia, yganjali}@cs.toronto.edu
![Page 2: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/2.jpg)
TCP, Oh TCP!
2
![Page 3: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/3.jpg)
TCP, Oh TCP!
๏ TCP congestion control
2
![Page 4: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/4.jpg)
TCP, Oh TCP!
๏ TCP congestion control
๏Focus on evolution of cwnd over RTT.
2
![Page 5: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/5.jpg)
TCP, Oh TCP!
๏ TCP congestion control
๏Focus on evolution of cwnd over RTT.
๏ Damages
2
![Page 6: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/6.jpg)
TCP, Oh TCP!
๏ TCP congestion control
๏Focus on evolution of cwnd over RTT.
๏ Damages
๏ TCP pacing
2
![Page 7: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/7.jpg)
The Tortoise and the Hare: Why Bother Pacing?
3
![Page 8: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/8.jpg)
The Tortoise and the Hare: Why Bother Pacing?
๏ Renewed interest in pacing for the data center environment๏Small buffer switches
๏Small round-trip times
๏Disparity between the total capacity of the network and the capacity of individual queues
3
![Page 9: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/9.jpg)
The Tortoise and the Hare: Why Bother Pacing?
๏ Renewed interest in pacing for the data center environment๏Small buffer switches
๏Small round-trip times
๏Disparity between the total capacity of the network and the capacity of individual queues
๏Focus on tail latency cause by short-term
unfairness in TCP
3
![Page 10: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/10.jpg)
TCP Pacing’s Potential
4
![Page 11: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/11.jpg)
TCP Pacing’s Potential
๏ Better link utilization on small switch buffers
4
![Page 12: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/12.jpg)
TCP Pacing’s Potential
๏ Better link utilization on small switch buffers๏ Better short-term fairness among flows of
similar RTTs:๏ Improves worst-flow latency
4
![Page 13: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/13.jpg)
TCP Pacing’s Potential
๏ Better link utilization on small switch buffers๏ Better short-term fairness among flows of
similar RTTs:๏ Improves worst-flow latency
๏ Allows slow-start to be circumvented๏Saving many round-trip time๏May allow much larger initial congestion window to
be used safely
4
![Page 14: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/14.jpg)
Contributions
5
![Page 15: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/15.jpg)
Contributions
๏ Effectiveness of TCP pacing in data centers.
5
![Page 16: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/16.jpg)
Contributions
๏ Effectiveness of TCP pacing in data centers.
๏ Benefits of using paced TCP diminish as we increase
the number of concurrent connections beyond a
certain threshold (Point of Inflection).
5
![Page 17: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/17.jpg)
Contributions
๏ Effectiveness of TCP pacing in data centers.
๏ Benefits of using paced TCP diminish as we increase
the number of concurrent connections beyond a
certain threshold (Point of Inflection).
๏ Inconclusive results in previous works.
5
![Page 18: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/18.jpg)
Contributions
๏ Effectiveness of TCP pacing in data centers.
๏ Benefits of using paced TCP diminish as we increase
the number of concurrent connections beyond a
certain threshold (Point of Inflection).
๏ Inconclusive results in previous works.
๏ Inter-flow bursts.
5
![Page 19: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/19.jpg)
Contributions
๏ Effectiveness of TCP pacing in data centers.
๏ Benefits of using paced TCP diminish as we increase
the number of concurrent connections beyond a
certain threshold (Point of Inflection).
๏ Inconclusive results in previous works.
๏ Inter-flow bursts.
๏ Test-bed experiments.
5
![Page 20: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/20.jpg)
Inter-flow Bursts
6
![Page 21: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/21.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity
6
![Page 22: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/22.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size
6
![Page 23: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/23.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.
6
![Page 24: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/24.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.
6
![Page 25: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/25.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.
6
0 RTT 2RTT 3RTT
![Page 26: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/26.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.
6
0 RTT 2RTT 3RTT
![Page 27: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/27.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.
6
0 RTT 2RTT 3RTT
![Page 28: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/28.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.
6
0 RTT 2RTT 3RTT
![Page 29: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/29.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.๏ X: Inter-flow burst
6
0 RTT 2RTT 3RTT
![Page 30: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/30.jpg)
Inter-flow Bursts
๏ C: bottleneck link capacity๏ Bmax :buffer size๏ N: longed lived flows.๏ W: packets in every RTT in paced or non-paced
manner.๏ X: Inter-flow burst
6
0 RTT 2RTT 3RTT
![Page 31: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/31.jpg)
Modeling
7
![Page 32: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/32.jpg)
Modeling
7
best case of non-paced
![Page 33: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/33.jpg)
Modeling
7
best case of non-paced worst case of paced
![Page 34: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/34.jpg)
Modeling
7
RTT
best case of non-paced worst case of paced
![Page 35: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/35.jpg)
Modeling
7
RTT RTT
best case of non-paced worst case of paced
![Page 36: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/36.jpg)
Modeling
7
RTT RTT
best case of non-paced worst case of paced
![Page 37: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/37.jpg)
Modeling
7
RTT RTT
best case of non-paced worst case of paced
![Page 38: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/38.jpg)
Experimental Studies
8
![Page 39: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/39.jpg)
Experimental Studies
๏ Flow of sizes 1,2, 3 MB between servers and clients.
8
![Page 40: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/40.jpg)
Experimental Studies
๏ Flow of sizes 1,2, 3 MB between servers and clients.๏ Bottleneck BW: 1,2, 3 Gbps
8
![Page 41: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/41.jpg)
Experimental Studies
๏ Flow of sizes 1,2, 3 MB between servers and clients.๏ Bottleneck BW: 1,2, 3 Gbps๏ RTT: 1 to 100ms
8
![Page 42: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/42.jpg)
Experimental Studies
๏ Flow of sizes 1,2, 3 MB between servers and clients.๏ Bottleneck BW: 1,2, 3 Gbps๏ RTT: 1 to 100ms๏ Bottleneck utilization, Drop rate, average and tail
FCT
8
![Page 43: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/43.jpg)
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
![Page 44: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/44.jpg)
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
No congestion
![Page 45: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/45.jpg)
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
No congestion Congestion
![Page 46: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/46.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
No congestion Congestion
![Page 47: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/47.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
No congestion Congestion
![Page 48: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/48.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
38%
No congestion Congestion
![Page 49: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/49.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
38%
No congestion Congestion
![Page 50: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/50.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
9
38%
No congestion Congestion
![Page 51: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/51.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
90.1 0.2 0.3 0.40.0630.0390
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
38%
No congestion Congestion
![Page 52: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/52.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
90.1 0.2 0.3 0.40.0630.0390
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
38%
No congestion Congestion
1RTT
![Page 53: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/53.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
90.1 0.2 0.3 0.40.0630.0390
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
38%
No congestion Congestion
1RTT2RTTs
![Page 54: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/54.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
90.1 0.2 0.3 0.40.0630.0390
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
1 2 3 50.06 70.50.20
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
38%
No congestion Congestion
1RTT2RTTs
![Page 55: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/55.jpg)
0 50 100 150 200 250 3000
50
100
150
200
250
300
Time (sec)
Bottle
neck
Lin
k U
tiliz
atio
n (
Mbps)
pacednon!paced
Base-Case Experiment: One flow vs Two flows, 64KB of buffering, Utilization/Drop/FCT
90.1 0.2 0.3 0.40.0630.0390
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
1 2 3 50.06 70.50.20
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Flow Completion Time (sec)
CD
F
pacednon!paced
38%
No congestion Congestion
1RTT2RTTs 2RTTs
![Page 56: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/56.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
10
![Page 57: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/57.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
10
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
paced
non!paced
PoIN*
![Page 58: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/58.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
10
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
paced
non!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Bo
ttle
ne
ck L
ink
Dro
p (
%)
pacednon!paced
PoIN*
![Page 59: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/59.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
10
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
pacednon!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
paced
non!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Bo
ttle
ne
ck L
ink
Dro
p (
%)
pacednon!paced
PoIN*
![Page 60: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/60.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
100 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of Flows
99
th P
erc
en
tile
FC
T (
sec)
pacednon!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
pacednon!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
paced
non!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Bo
ttle
ne
ck L
ink
Dro
p (
%)
pacednon!paced
PoIN*
![Page 61: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/61.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 1.7% of BDP, varying number of flows
100 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of Flows
99
th P
erc
en
tile
FC
T (
sec)
pacednon!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
pacednon!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
paced
non!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Bo
ttle
ne
ck L
ink
Dro
p (
%)
pacednon!paced
PoIN*
• Number of concurrent connections increase beyond a certain point the benefits of pacing diminish.
![Page 62: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/62.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 3.4% of BDP, varying number of flows
11
![Page 63: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/63.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 3.4% of BDP, varying number of flows
110 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of Flows
99
th P
erc
en
tile
FC
T (
sec)
paced
non!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
paced
non!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*
PoI
0 20 40 60 80 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Number of flows sharing the bottleneck
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 64: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/64.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 3.4% of BDP, varying number of flows
110 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of Flows
99
th P
erc
en
tile
FC
T (
sec)
paced
non!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
paced
non!paced
N*
PoI
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*
PoI
0 20 40 60 80 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Number of flows sharing the bottleneck
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
• Aagarwal et al.: Don’t pace!
• 50 flows, BDP 1250 packets and buffer size
312 packets
• N* = 8 flows.
• Kulik et al.: Pace!
• 1 flow, BDP 91 packets, buffer size 10
packets.
• N* = 9 flows.
![Page 65: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/65.jpg)
N* vs. Buffer
12
![Page 66: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/66.jpg)
50 100 150 200 2500
200
400
600
800
1000
Buffer Size (KB)
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N* vs. Buffer
12
![Page 67: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/67.jpg)
50 100 150 200 2500
200
400
600
800
1000
Buffer Size (KB)
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N* vs. Buffer
1250 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Buffer size(KB)
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 68: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/68.jpg)
50 100 150 200 2500
200
400
600
800
1000
Buffer Size (KB)
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
50 100 150 200 2500
0.2
0.4
0.6
0.8
1
1.2
1.4
Buffer Size (KB)
Ave
rag
e �C
T (
se
c)
pacednon!paced
N* vs. Buffer
1250 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Buffer size(KB)
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 69: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/69.jpg)
50 100 150 200 2500
0.4
0.8
1.2
1.6
2
2.4
2.8
Buffer Size (KB)
99
th P
erc
en
tile
�CT
(se
c)
pacednon!paced
50 100 150 200 2500
200
400
600
800
1000
Buffer Size (KB)
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
50 100 150 200 2500
0.2
0.4
0.6
0.8
1
1.2
1.4
Buffer Size (KB)
Ave
rag
e �C
T (
se
c)
pacednon!paced
N* vs. Buffer
1250 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Buffer size(KB)
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 70: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/70.jpg)
50 100 150 200 2500
0.4
0.8
1.2
1.6
2
2.4
2.8
Buffer Size (KB)
99
th P
erc
en
tile
�CT
(se
c)
pacednon!paced
50 100 150 200 2500
200
400
600
800
1000
Buffer Size (KB)
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
50 100 150 200 2500
0.2
0.4
0.6
0.8
1
1.2
1.4
Buffer Size (KB)
Ave
rag
e �C
T (
se
c)
pacednon!paced
N* vs. Buffer
1250 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Buffer size(KB)
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 71: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/71.jpg)
Clustering Effect: The probability of packets from a flow being followed by packets from other flows
13
![Page 72: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/72.jpg)
Clustering Effect: The probability of packets from a flow being followed by packets from other flows
Non-paced: Packets of each flow are clustered together.
13
![Page 73: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/73.jpg)
Clustering Effect: The probability of packets from a flow being followed by packets from other flows
Non-paced: Packets of each flow are clustered together.
Paced: Packets of different flows are multiplexed.
13
![Page 74: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/74.jpg)
Drop Synchronization: Number of Flows Affected by Drop Event
14
![Page 75: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/75.jpg)
Drop Synchronization: Number of Flows Affected by Drop Event
14
NetFPGA router to count the number of flows affected by drop events.
![Page 76: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/76.jpg)
Drop Synchronization: Number of Flows Affected by Drop Event
14
0 10 20 30 400
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
0 20 40 60 800
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
0 50 100 150 200 250 300 350 4000
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
N: 48 N: 96 N: 384
NetFPGA router to count the number of flows affected by drop events.
![Page 77: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/77.jpg)
Drop Synchronization: Number of Flows Affected by Drop Event
14
0 10 20 30 400
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
0 20 40 60 800
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
0 50 100 150 200 250 300 350 4000
0.10.20.30.40.50.60.70.80.9
1
Number of Flows Affected by Drop Event
CD
F
paced
non!paced
N: 48 N: 96 N: 384
NetFPGA router to count the number of flows affected by drop events.
![Page 78: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/78.jpg)
Future Trends for Pacing: per-egress pacing.
15
![Page 79: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/79.jpg)
Future Trends for Pacing: per-egress pacing.
15
6 12 24 48 96 1920
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e R
CT
(se
c)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
0.20.40.60.8
11.21.41.61.8
22.2
Number of Flows
99
th P
erc
en
tile
RC
T (
sec)
per!flow pacednon!pacedper!host + per!flow paced
N*PoI
6 12 24 48 96 1920
5
10
15
20
Number of flows
Bottle
neck
Lin
k D
rop (
%)
per!flow pacednonpacedper!host + per!flow paced
![Page 80: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/80.jpg)
Future Trends for Pacing: per-egress pacing.
15
6 12 24 48 96 1920
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e R
CT
(se
c)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
0.20.40.60.8
11.21.41.61.8
22.2
Number of Flows
99
th P
erc
en
tile
RC
T (
sec)
per!flow pacednon!pacedper!host + per!flow paced
N*PoI
6 12 24 48 96 1920
5
10
15
20
Number of flows
Bottle
neck
Lin
k D
rop (
%)
per!flow pacednonpacedper!host + per!flow paced
![Page 81: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/81.jpg)
Future Trends for Pacing: per-egress pacing.
15
6 12 24 48 96 1920
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e R
CT
(se
c)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
0.20.40.60.8
11.21.41.61.8
22.2
Number of Flows
99
th P
erc
en
tile
RC
T (
sec)
per!flow pacednon!pacedper!host + per!flow paced
N*PoI
6 12 24 48 96 1920
5
10
15
20
Number of flows
Bottle
neck
Lin
k D
rop (
%)
per!flow pacednonpacedper!host + per!flow paced
![Page 82: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/82.jpg)
Future Trends for Pacing: per-egress pacing.
15
6 12 24 48 96 1920
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e R
CT
(se
c)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
0.20.40.60.8
11.21.41.61.8
22.2
Number of Flows
99
th P
erc
en
tile
RC
T (
sec)
per!flow pacednon!pacedper!host + per!flow paced
N*PoI
6 12 24 48 96 1920
5
10
15
20
Number of flows
Bottle
neck
Lin
k D
rop (
%)
per!flow pacednonpacedper!host + per!flow paced
![Page 83: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/83.jpg)
Future Trends for Pacing: per-egress pacing.
15
6 12 24 48 96 1920
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e R
CT
(se
c)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
per!flow pacednon!pacedper!host + per!flow paced
PoIN*
6 12 24 48 96 1920
0.20.40.60.8
11.21.41.61.8
22.2
Number of Flows
99
th P
erc
en
tile
RC
T (
sec)
per!flow pacednon!pacedper!host + per!flow paced
N*PoI
6 12 24 48 96 1920
5
10
15
20
Number of flows
Bottle
neck
Lin
k D
rop (
%)
per!flow pacednonpacedper!host + per!flow paced
![Page 84: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/84.jpg)
Conclusions and Future work
๏ Re-examine TCP pacing’s effectiveness:๏Demonstrate when TCP pacing brings benefits in
such environments.๏ Inter-flow burstiness
๏ Burst-pacing vs. packet-pacing.๏ Per-egress pacing.
16
![Page 85: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/85.jpg)
Renewed Interest
17
![Page 86: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/86.jpg)
Traffic Burstiness Survey
๏‘Bursty’ is a word with no agreed meaning. How do you define a bursty traffic?
๏If you are involved with a data center, is your data center traffic bursty?
๏ If yes, do you think that it will be useful to supress the burstiness in your traffic?
๏ If no, are you already supressing the burstiness? How? Would you anticipate the traffic becoming burstier in the future?
18
![Page 87: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/87.jpg)
19
![Page 88: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/88.jpg)
Base-Case Experiment: One RPC vs Two RPCs, 64KB of buffering, Latency
20
![Page 89: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/89.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size: 6% of BDP, varying number of
21
![Page 90: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/90.jpg)
Base-Case Experiment: One RPC vs Two RPCs, 64KB of buffering, Latency / Queue Occupancy
22
![Page 91: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/91.jpg)
Base-Case Experiment: One RPC vs Two RPCs, 64KB of buffering, Latency / Queue Occupancy
22
![Page 92: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/92.jpg)
Base-Case Experiment: One RPC vs Two RPCs, 64KB of buffering, Latency / Queue Occupancy
22
![Page 93: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/93.jpg)
Base-Case Experiment: One RPC vs Two RPCs, 64KB of buffering, Latency / Queue Occupancy
22
![Page 94: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/94.jpg)
Functional test
23
![Page 95: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/95.jpg)
Functional test
23
![Page 96: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/96.jpg)
Functional test
23
![Page 97: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/97.jpg)
Functional test
23
![Page 98: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/98.jpg)
RPC vs. Streaming
24
1GE10GE 10GE
RTT = 10ms
Paced by ack clocking
Bursty
![Page 99: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/99.jpg)
Zooming in more on the paced flow
25
![Page 100: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/100.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 6.8% of BDP, varying number of flows
26
![Page 101: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/101.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 6.8% of BDP, varying number of flows
26
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*PoI
![Page 102: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/102.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 6.8% of BDP, varying number of flows
26
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*PoI
0 20 40 60 80 100−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Number of flows sharing the bottleneck
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 103: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/103.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 6.8% of BDP, varying number of flows
26
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
pacednon!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*PoI
0 20 40 60 80 100−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Number of flows sharing the bottleneck
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced
![Page 104: TCP Pacing in Data Center Networks](https://reader031.vdocuments.pub/reader031/viewer/2022021508/58499aa81a28aba93a9165a8/html5/thumbnails/104.jpg)
Multiple flows: Link Utilization/Drop/LatencyBuffer size 6.8% of BDP, varying number of flows
260 10 20 30 40 50 60 70 80 90 1000
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
Number of Flows
99
th P
erc
en
tile
FC
T (
sec)
pacednon!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
Number of Flows
Ave
rag
e F
CT
(se
c)
pacednon!paced
PoIN*
0 10 20 30 40 50 60 70 80 90 1000
200
400
600
800
1000
Number of Flows
Bo
ttle
ne
ck L
ink
Util
iza
tion
(M
bp
s)
pacednon!paced
N*PoI
0 20 40 60 80 100−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Number of flows sharing the bottleneck
Bottl
enec
k lin
k dr
op(%
)
pacednonpaced