Download - FAST TCP : From Theory to Experiments
![Page 1: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/1.jpg)
FAST TCP: From Theory to Experiments
Sohn JongSooIntelligent Information System Lab.
![Page 2: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/2.jpg)
Index Introduction
› Motivation› Background theory› Equilibrium› Stability
FAST TCP Parameter calculation Experiments Conclusion
![Page 3: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/3.jpg)
Introduction Congestion control algorithm
› Size, speed, load, connectivity› Current TCP implementation
Performance bottleneck Different congestion control algorithm for TCP
› Called FAST TCP› Equation-based algorithm
Eliminating packet-level oscillations› Using queuing delay as the primary measure of conges-
tion› Weighted proportioanl fairness in equilibrium
Does not penalize long flows, as the current congestions control algorithm
![Page 4: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/4.jpg)
Motivation High Energy and Nuclear Physics
(HENP)› 2,000 physicists, 150 universities, 30 coun-
tries› Require ability to analyze and share many
terabyte-scale data collections Key Challenge
› Current congestion control algorithm of TCP does not scale to this regime
4
![Page 5: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/5.jpg)
Background Theory Congestion control consists of:
› Source algorithm (TCP) that adapts send-ing rate (window) based on congestion feedback
› Link algorithm (at router) that updates and feeds back a measure of congestion
› Typically link algorithm is implicit and mea-sure of congestion is loss probability or queueing delay
5
![Page 6: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/6.jpg)
Background Theory (cont.) Preliminary theory studies equilibrium
and stability properties of the source-link algorithm pair› Source-link algorithm is TCP/AQM (active
queue management)
6
![Page 7: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/7.jpg)
Equilibrium Interpret TCP/AQM as a distributed al-
gorithm to solve a global optimization problem
Can be broken into two sub-problems› Source (maximize sum of all data rates)› Link (minimize congestion)
7
![Page 8: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/8.jpg)
Equilibrium (cont.) Source
› Each source has a utility function, as a function of its data rate
› Optimization problem is maximizing sum of all utility functions over their rates
› Challenge is solving for optimal source rates in a distributed manner using only local information
8
![Page 9: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/9.jpg)
Equilibrium (cont.) Exploit Duality Theory
› Associated with primal (source) utility max-imization is dual (link congestion) mini-mization problem
› Solving the dual problem is equivalent to solving the primal problem
› Class of optimization algorithms that itera-tively solve for both at once
9
![Page 10: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/10.jpg)
Stability Want to ensure equilibrium points are
stable› When the network is perturbed out of equi-
librium by random fluctuations, should drift back to new equilibrium point
Current TCP algorithms can become unstable as delay increases or network capacity increases
10
![Page 11: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/11.jpg)
Lack of Scalability of TCP As capacity in-
creases, link uti-lization of TCP/RED steadily drops
Main factor may be synchroniza-tion of TCP flows
11
ns-2 simulation
capacity = 155Mbps, 622Mbps, 2.5Gbps, 5Gbps, 10Gbps; 100 ms round trip latency; 100 flows
![Page 12: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/12.jpg)
FAST TCP Implementation issues
› Use both queueing delay and packet loss as congestion signals
› Effectively deal with massive losses› Use pacing to reduce burstiness and mas-
sive losses› Converge rapidly to neighbourhood of equi-
librium value after packet losses
12
![Page 13: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/13.jpg)
FAST TCP - Implementation
Congestion window update
13
W Window size
baseRTT Minimum observed RTT xi qi
Current throughput Current queuing delay
Protocol parameter
Weight = min(3/cwnd, 1/8)α = the number of pack-ets each flow buffered in the network
![Page 14: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/14.jpg)
Calculating parameters (cont.)
› Specifies total number of packets a single
FAST connection tries to maintain in queues along its path
› Can be used to tune the aggressiveness of the window update function
14
![Page 15: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/15.jpg)
Calculating parameters RTT and Wold
› Timestamp every packet stored in re-transmit queue and store Wcurrent
› On receipt of ACK, set RTT sample to dif -ference in times. Also, set Wold to the stored window size of the ACK’ed packet
baseRTT is set to minimum observed RTT sample
15
![Page 16: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/16.jpg)
Experiments - Infrastructure
16
![Page 17: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/17.jpg)
Infrastructure - Details
17
1, 2 Linux/FAST flows10,037 km
7, 9, 10 FAST flows3,948 km
![Page 18: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/18.jpg)
Throughput and Utilization
Statistics in parentheses are for cur-rent Linux TCP implementation
“bmps” = product of throughput and distance (bits-per-meter-per-second)
18
![Page 19: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/19.jpg)
Average Utilization Traces
19
![Page 20: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/20.jpg)
Traces (cont.)
20
![Page 21: FAST TCP : From Theory to Experiments](https://reader036.vdocuments.pub/reader036/viewer/2022062315/56815cf5550346895dcaf7eb/html5/thumbnails/21.jpg)
Conclusions Describing the development of FAST
TCP› Background theory to actual implementa-
tion High utilization
› Without having to fill the buffer and incur large queuing delay
FAST TCP can converge rapidly› Yet stably