1

TCP Schemes Investigation in TCP Schemes Investigation in Wired and Wireless Hybrid Wired and Wireless Hybrid

NetworksNetworks

By Weiwei Hu and Wei Zha

Dec. 8, 2004

2

Outline Introduction Random Loss Model TCP-Jersey TCP-Westwood Performance Analysis Conclusion References

3

Introduction

Transmission Control Protocol (TCP) reliable end-to-end window based designed for the wired networks (no random packet losses): not fit for

wireless network

Application WWW - HTTP File transfer - FTP E-Mail - SMTP Remote access - Telnet

4

Introduction (Cont.)

Wireless networks: bursty and high channel error rates.

Throughput decreases greatly.

Extending TCP as used over the wired links to the wireless links may bring inefficient usage of bandwidth.

The TCP protocol is still used to transfer data over the wireless link.

5

Introduction (cont.) A lot of attention is currently being paid to the

design of a better TCP scheme over wireless links.

Difficulty in modeling the TCP protocol analytically

Many of these studies are simulation based.

Hard to obtain insight into the effects of particular parameters on the behavior of TCP using simulations with specific settings

6

Random Loss Model Independent identically distributed (i.i.d )

The packet transmission time of the packets on both lossless and

lossy link are exponentially distributed. The congestion avoidance phrase is modeled using probability.

Correlated Hiding the losses from the transport layer and moving the problem

to link layer Applicable on very low bandwidth wireless links cannot be used to model the fast-transmit and fast-recovery phrases

7

TCP-Jersey

Congestion warning marking function (RED and ECN)

Routers can take part in controlling TCP transmission rate: active queue management (AQM)

Random early detection (RED): AQM scheme implementation Explicit congestion notification (ECN): an extension to RED RED and ECN are sensitive to parameter settings

Cause unsatisfactory TCP performance

8

TCP-Jersey (Cont.) TCP-Jersey simplifies the marking function:

use only one threshold (thresh) Marking all the packets with probability 1 if the

average queue length exceeds the predefined threshold.

Calculating the average queue length using a larger queue weight than the original ECN.

9

TCP-Jersey (Cont.) Available Bandwidth Estimation Estimate the available bandwidth (Rn) for the TCP

connection. ABE estimator:

Calculate the optimum congestion windowoptimum congestion window (ownd) size once every RTT.

RTTtt

LRRTTR

nn

nnn

)( 1

1

sizeseg

RRTTownd n

n _

Ln: size of packet n;

t n-1: previous packet arrival time.

10

TCP-Westwood

A sender-side modification of the TCP congestion window algorithm

improves upon the performance of TCP in wired and mixed network

uses an end-to-end bandwidth measurement. The ACK-based measurement procedure filtering the ACK reception rate The effects of delayed and cumulative ACKs on bandwidth

measurement

11

TCP-Westwood (Cont.)

Main idea: exploit TCP acknowledgement packets to derive rather complicated measurements.

A source perform an end-to-end estimate of the bandwidth available along a TCP connection by measuring and averaging the rate of returning ACKs.

1

kk

kk tt

db

12

12

12

1

11

1

1

kk

kkk

kk

kkk

tt

bbb

tt

ttb

12

TCP-Westwood (Cont.)

Three types of filters actually used in the simulation: Original filter:

current_bwe_ = current_bwe * fr_alpha_ + sample_bwe * (1 – fr_alpha_)

Filter 1: current_bwe_ = current_bwe_ * .9047 + (sample_bwe + last_bwe_sample_) * .0476

Filter 2: (LPF)current_bwe_ = cureent_bwe_ * .93548 + (sample_bwe+ last_bwe_sample_) * .03225

][2

)1(11

kkk bba

bab

13

TCP-Westwood (Cont.) TCP-Westwood controls the window using end-to-end rate

transmission to continuously estimate the packet rate of the connection at the TCP sender.

It can reset the window to match available bandwidth, which makes it more robust to the random loss in wireless link.

TCP-Westwood has many superior features which can handle losses caused by link errors or wireless channel conditions more efficiently than most of previous existing TCP schemes.

14

Performance Analysis

We chose TCP-Westwood as a starting point because of its similarity to TCP-Jersey.

We tried to simulate TCP-Jersey in ns2. However, due to the opaque description for detailed mechanism in TCP-Jersey and lack of references, the implementation seems infeasible so far.

Problem: The congestion warning part (how to define the proper thresh of average queue length).

Thus, we simulate TCP-Westwood and all the other conventional TCP schemes in NS-2 network simulator.

15

Simulation Environment

The source connects to a wired based station via a 20 Mb/s error free link with 35 ms one-way delay.

The base station is linked to the destination, a wireless mobile node, via a 2 Mb/s lossy channel with 1 ms delay.

A single TCP connection running a long-live FTP application delivers data from the source to the destination.

S BS D20Mb, 35ms

2 Mb, 1 ms

16

Bandwidth Estimation

An effective way to calculate the congestion window size

Control data traffic efficiently.

Expecting TCP-Westwood can reach a good estimation for bandwidth so that the lossy link can be utilized as much as possible.

17

Bandwidth Estimation (Cont.)

• When the simulation time is greater than 7s, the TCP Westwood reaches the max bandwidth utilization.

18

Bandwidth Estimation (Cont.)

• The slow start threshold matches very well with congestion window settings.

19

Goodput

An explicit characteristic for TCP performance. The effective amount of data delivered through the

network. Run the simulation for TCP -Tahoe, -Reno, New

Reno, -Sack, and -Westwood respectively. Random link error rate varies from 0.001% to

10%. The simulation time is 60s.

20

Goodput Performance (Cont.)

10-5

10-4

10-3

10-2

10-1

200

400

600

800

1000

1200

1400

1600

1800

Goo

dput

(K

bps)

Wireless Link Error Rate

TahoeRenoNew RenoSackWestwood

• The first point at link error rate 0.00001, all TCP schemes are closely match to each other, which means that they perform the same at nearly no loss rate condition.

• When the lossy rate increases, TCP Westwood outperforms other TCP schemes explicitly, especially when error rate increases to 0.001.

21

Goodput Performance (Cont.)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

Goo

dput

(M

bps)

Bottleneck bandwidth (Mbps)

TahoeRenoNew RenoSackWestwood• The buffer size is kept

as a constant, B = 60pkts.

• TCP-Westwood varies almost linearly to track the bandwidth variations of the bottleneck.

• TCP-Westwood is a slightly better than other schemes.

22

Packet Loss

3rd dup Ack triggles a

retransmission of

packet #115000:

Fast Recovery

23

Packet Loss (cont.)

The sending rate is slowed down:

Congestion Avoidance phase

24

Packet Loss (cont.)

Reenter the Slow Start phase

25

Conclusion Investigate the performance of the different TCP algorithm

in a wired and wireless hybrid networks. Provide an analytical model for TCP-Jersey and TCP-

Westwood. First, we started with TCP-Westwood to perform the

simulation. Haven’t reproduced the TCP-Jersey so far in results of

obscure description and lack of references. Limitation:

Congestion Warning: restrict the threshold of marking algorithm. Cooperate ECN to mark the packets.

26

Conclusion (cont.)

Different TCP algorithms, namely, Tahoe, Reno, New-Reno, Sack, and Westwood are studied in the given simulation environment. According to the experiment results, TCP-Westwood operates more robust than other existing TCP schemes, which shows the benefits of the bandwidth estimation in wireless or hybrid networks.

27

References

1) Kai Xu, Ye Tian, and Nirwan Ansari, “TCP-Jersey for Wireless IP Communications”, IEEE Journal on Selected areas in Communications, vol. 22, no.4, May 2004.

2) Gerla M., Sanadidi M.Y., Ren Wang Zanella, A. Casetti C., and Mascolo S., “TCP Westwood: Bandwidth Estimation for Enhanced Transport over Wireless Links”, IEEE Global Telecommunications Conference 2001, vol. 3,  pp. 25-29, Nov. 200.

3) David D. Clark, and Wenjia Fang, “Explicit Allocation of Best-Effort Packet Delivery Service”, IEEE/ACM Trans. Networking, vol. 6, pp.362-373.

4) Farooq Anjum and Leandros Tassiulas, “Comparative study of various TCP versions over a wireless link with correlated losses”, IEEE/ACM Transaction on Networking, vol. 11, no. 3, June 2003.

5) A. Kumar and J. Holltzmann, “Comparitive performance analysis of versions of TCP in local network with a lossy link – Part II: Rayleigh Fading Mobile Radio Link”, Rutgers Univ., New Brunswick, NJ, Tech. Rep. WINLAB-TR 129, Oct. 1996.

6) M.May, J. Bolot, C. Diot, and B. Lyles, “Reasons not to deploy FED,” Proc. Int. Workshop Quality-of-Service (IWQoS), 1999, pp. 260-262.

7) A. Chochalingam, M. Zozi, and R. R. Rao, “Performance of TCP on wireless fading links with memory,” in Proc. IEEE Int. Conf. Communications, June 1998, pp. 595-600.

28

References (Cont.)8) V. Jacobson, “Congestion avoidance and control,” ACM Computer Communications

Review, vol. 4, pp. 314-319, Auguest 1988.9) C. Casetti, M. Gerla, S. Lee, S. Mascolo, and M. Sanadidi, “TCP with fast recovery”,

MILCOM 2000, Los Angeles, CA, Oct. 2000.10) A. Zanella, G. Procissi, M. Gerla, M. Y. Snanaddi, “TCP Westwood: analytic model

and performance evaluation”, Technical Report, URL: http://www.cs.ucla.edu/csd/phus/pubs.html.

11) ns-2 network simulator. LBL, URL: http://www.msh.cs.berkley.edu/ns.12) TCP Westwood modules for ns-2. URL:

http://www1.tcl.polito.it/casetti/tco-westwood.13) K. Fall and S. Floyd, “Simulation-based comparisons of Tahoe, Reno, and SACK

TCP,” ACM Computer Communication Review, 16:5-21, 1996.14) C. Casetti and M. Meo, “A new approach to model the stationary havavior of TCP

connections”, In Proc. of IEEE INFOCOME, pp. 367-375, Mar. 2000.15) J. Mo, R. La, V. Anantharam, and J. Walrand, “Analysis and comparision

of TCP Reno and Vegas,” In Proc. of IEEE INFOCOM, pp. 1556-1563, Mar. 1999.16) J. Padhye, V. Firoiu, D. Towsley, and J. Kurose, “Modeling TCP throughput: a

simple model and its empirical validation,” ACM Computer Communication Review, 28:303-314.