Top Banner
TCP Friendliness CMPT771 Spring 2008 Michael Jia
22
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

TCP Friendliness

CMPT771

Spring 2008

Michael Jia

Page 2: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Outline

Background

Classification

Achievements

Challenges

Page 3: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

TCP Fairness

R

R

equal bandwidth share

Connection 1 throughputConnect

ion 2

th

roughput

Fair: 1. Equal share

2. Full utilizationif K TCP sessions share same bottleneck link of bandwidth R, each should have average rate of R/K

Page 4: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Why UDP?

UDP Preferred Applications• Video Streaming• VoIP

UDP Advantages• Simplicity • Lower overhead (light weight)• No re-transmission required

Page 5: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Problem with UDP: Unresponsive Flows

• No congestion control• No response to packet drops• TCP competing with unresponsive UDP

– TCP flows reduce sending rates in response to congestion

– Uncooperative UDP flows capture the available bandwidth

– Unfair to TCP, or even starve TCP

Page 6: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Objective: TCP-friendly

TCP TCP

Internetnon-TCP

non-TCP

“ long-term throughput does not exceed the throughput of a conformant TCP connection under the same conditions”

Page 7: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Outline

Background

Classification

Achievements

Challenges

Page 8: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Classification

• Window-Based vs. Rate-Based– window-based:

• Window size controls rate• Sender or receiver(s)• Similar to TCP

– rate-based:• TCP throughput models• More smoother rate• Good for media streams

Page 9: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Classification

• Unicast vs. Multicast– Multicast: more difficult– RTT is required for Rate-based schemes– Window-based approach is more suitable

• Single-rate vs. Multi-rate– Unicast = Single-rate– Multicast: multi-rate protocols are preferred

• More flexible allocation of bandwidth• Layered multicast• Group management

Page 10: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Outline

Background

Classification

Achievements

Challenges

Page 11: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

TCP Throughput Equation 1R -- Bandwidth of TCP connection (Long term throughput)T -- Round-trip delay T (RTT)L --Packet size Lp -- Loss event rate p

T. Ott, J.H.B. Kemperman, M. Mathis, 1996

The Stationary Behavior of Ideal TCP Congestion Avoidance

Page 12: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

TCP Throughput Equation 2

22 3(3 ) (1 32 )

3 8RTO

LR

q qT T q q

=+ +

Padhye, J., Firoiu, V., Towsley, D., and Kurose, J., Modeling TCP Throughput: a Simple Model and its Empirical Validation, UMASS CMPSCI Tech Report TR98-008,

Feb. 1998.

R -- Bandwidth of TCP connection

T -- Round-trip delay T (RTT)

L --Packet size L

q -- Loss event rate q

TRTO -- Retransmission timeout (~ 4T)

Page 13: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

TCP Throughput Equation

M. Mathis, J. Semke, J. Mahdavi, and T. Ott. The macroscopic behavior of the TCP congestion avoidance algorithm.

Computer Communication Review, 27(3), July 1997

• Verify through simulation & live Internet measurements

• Assumption– Steady State (Ignore slow start phase & No timeouts)

– Constant packet size

Page 14: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Achievements

Page 15: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Achievements - TFRC

• TCP-Friendly Rate Control Protocol (2000)• Unicast, rate-based• Based on TCP equation 2

• Using more sophisticated methods to gather parameters– Average-Loss-Interval loss rate estimation

• Stable sending rate• Sufficient responsiveness

Page 16: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Achievements - TEAR

• TCP Emulation At Receivers (2000)• Multicast, single-rate• Rate-based + Window-based• Receiver maintains a congestion window• Receiver calculates average rate

– then send back to the sender– avoid saw-tooth-like behavior

• Scalable in multicast case– use the minimum rate

Page 17: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Achievements – Rainbow

• Rainbow (2000)• Multicast, multi-rate,

window-based• Digital-Fountain• Receivers individually

request each data packet• Routers process requests• Receiver controls

congestion• Limitation – router

supporting

Page 18: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Outline

Background

Classification

Achievements

Challenges

Page 19: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Challenges

• Lack of standard methods for comparison

• Fairness definitions for multicast

• Improvement of the models for TCP traffics

• How to treat short-lived flows

• Much more…

Page 20: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

References• Robert Denda Joerg Widmer and Martin Mauve, 2001, A

survey on tcp-friendly congestion control

• T. Ott, J.H.B. Kemperman, M. Mathis, 1996, The Stationary Behavior of Ideal TCP Congestion Avoidance

• Padhye, J., Firoiu, V., Towsley, D., and Kurose, J., 1998, Modeling TCP Throughput: a Simple Model and its Empirical Validation

• M. Mathis, J. Semke, J. Mahdavi, and T. Ott., 1997, The macroscopic behavior of the TCP congestion avoidance algorithm

• Jitendra Padhye Sally Floyd, Mark Handley and Joerg Widmer, 2000, Equation-based congestion control for unicast applications

• Volkan Ozdemir Injong Rhee and Yung Yi., 2000, Tear: Tcp emulation at receivers – flow control for multimedia streaming

Page 21: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Questions?

Page 22: TCP Friendliness CMPT771 Spring 2008 Michael Jia.

Thank You