performance enhancement of tfrc in wireless ad hoc networks mingzhe li, choong-soo lee, emmanuel...

34
Performance Enhancement of TFRC in Wireless Ad Hoc Networks Mingzhe Li, Choong-Soo Lee, Emmanuel Agu, Mark Claypool and Bob Kinicki Computer Science Department Worcester Polytechnic Institute Worcester, Massachusetts

Post on 21-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Performance Enhancement of TFRC in Wireless Ad Hoc Networks

Mingzhe Li, Choong-Soo Lee, Emmanuel Agu,

Mark Claypool and Bob Kinicki

Computer Science Department

Worcester Polytechnic Institute

Worcester, Massachusetts

DMS 2004 September 9, 2004

2

OutlineOutline

Introduction Background TFRC Performance over Wireless

Networks RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

3

IntroductionIntroduction The objective is improved support for streaming

multimedia applications over wireless networks. The TCP Friendly Rate Control protocol (TFRC) was

designed for wired networks. It can perform poorly over wireless networks.

The 802.11 MAC layer wireless protocol uses Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) and Request-to-Send/Clear-to-Send (RTS/CTS) to avoid frame collisions.

TFRC performance suffers from the contention delays and drops known as RTS/CTS jamming and RTS/CTS-induced congestion.

DMS 2004 September 9, 2004

4

IntroductionIntroduction This paper introduces a wireless extension to

the TFRC protocol, Rate Estimation TFRC (RE-TFRC), that accounts for MAC layer saturation to select a sending rate that outperforms TFRC.

The goal of RE-TFRC is to reduce MAC layer loss rates and collisions and thereby lower transport layer delays with minimal effect on throughput.

DMS 2004 September 9, 2004

5

OutlineOutline

Introduction Background TFRC Performance over Wireless

Networks RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

6

TCP Friendly Rate Control (TFRC) [Floyd00]– Designed for streaming media applications– Uses rate-based congestion control and– The TCP Friendly congestion response function:

TFRC is implemented in the Linux kernel as one of the congestion control options of the Datagram Congestion Control Protocol (DCCP).

TCP Friendly Rate Control (TFRC)TCP Friendly Rate Control (TFRC)

83

)321((33

2 2 bpptp

bpr

sX

rto

X: Transmission rate s: packet sizer: round trip time p: lost event ratetrto:: Retransmission time out b: num of packets in each ack

DMS 2004 September 9, 2004

7

Hidden Terminal ProblemHidden Terminal Problem

Node 1 is hidden from Node 3:– Node 1 and node 3 cannot sense each other’s

transmissions.– If Node 1 and node 3 transmit at the same time to

node 2, a collision occurs at node 2.– Node 1 and node 3 back off and retransmit.

1 2 3

DMS 2004 September 9, 2004

8

Hidden Terminal ProblemHidden Terminal Problem

802.11 Solution to the Hidden Terminal Problem– Use a four-way handshake: RTS-CTS-DATA-ACK where the

RTS and CTS packets are significantly smaller than the average data packet.

– The maximum number of RTS retransmissions is set to 7.

However, the 802.11 protocol will still have problems if the MAC layer becomes saturated!!

1 2 3

DMS 2004 September 9, 2004

9

MAC Layer SaturationMAC Layer Saturation

MAC layer congestion– The wireless network traffic load is increased

above the MAC layer saturation point. – Contention delays and drops are increased.– The RTS/CTS jamming is hidden from upper layers.

TFRC then computes an ineffective RTT (Round Trip Time) and loss event rate.

This implies a TCP Friendly sending rate that is too high for optimal performance.

DMS 2004 September 9, 2004

10

OutlineOutline

Introduction Background TFRC Performance over Wireless

Networks RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

11

TFRC Performance InvestigationTFRC Performance Investigation NS-2 simulations are used. Evaluate a single flow, 802.11b MAC layer protocol over a chain topology

with a 2 Mbps wireless capacity. The throughput decreases as the number of hops increases.

DMS 2004 September 9, 2004

12

Rate Constrained TFRCRate Constrained TFRC A seven-hop chain network was simulated. The TFRC sending rate is manually constrained. The MAC layer saturates at 300Kbps.

DMS 2004 September 9, 2004

13

Rate Constrained TFRCRate Constrained TFRC The TFRC loss event rate and RTT increase sharply

after a 300Kbps constrained sending rate. Thus, unconstrained TFRC runs in a sub-optimal

state due to MAC layer congestion.

DMS 2004 September 9, 2004

14

OutlineOutline

Introduction Background TFRC Performance over Wireless

Network RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

15

Rate Estimation TFRC (RE-TRFC)Rate Estimation TFRC (RE-TRFC)

RE-TFRC estimates the optimum sending rate based on:

– The number of hops in the flow path– The current loss event rate.

The TFRC sending rate is adjusted depending on the estimate of the optimum sending rate.

RE-TFRC preserves the ceiling imposed by the TCP Friendly sending rate.

DMS 2004 September 9, 2004

16

Rate EstimationRate Estimation

Rate Estimate in TCP Westwood [wang02]– Upon congestion, Westwood sets the TCP

window size to

W = Bit-rateest * rttmin

– rttmin is the smallest recorded rtt, i.e., an estimate of latency.

RE-TRFC Rate Estimate Approach– Estimate the optimum sending rate that will not

saturate the MAC layer.

– Determine the MAC layer saturation rtt: rttopt

– Control the sending rate on congestion.

DMS 2004 September 9, 2004

17

TCP Friendly equation:

Inverse TCP function:

X: TCP Friendly rate

p: TFRC loss event rate

RE-TFRC Rate EstimationRE-TFRC Rate Estimation

Use R to estimate p’

Use p’ to estimate R’

),( prttfX

),( Xrttfp

),(' Rrttfp opt

)',(' prttfR cur

R: TFRC estimated receiving rate

p’: Adjusted TFRC loss event rate

R’: Estimated optimum sending rate

DMS 2004 September 9, 2004

18

Single hop delay model: [Carvalho03]

Multi-hop chain delay model:– Divide the N-hop chain into N-2 4-node networks

and two 3-node networks.– Sum the data/ack packet delay over the N hops.

Round Trip Time ModelingRound Trip Time Modeling

timebackoff average :

timensmission packet tra :

B

s

sB

T

t

tTT

DMS 2004 September 9, 2004

19

Round Trip Time ModelingRound Trip Time Modeling

1 2 3 4 5

)4()2()3(2)4()2()3(2

00

ackackdatadata

N

iack

N

idata

TNTTNT

TTNrii

: estimate of rttopt for N-hop chain topology

: Single hop delay of Ack packet

: Single hop delay of Data packet

)(Nr

ackT

dataT

DMS 2004 September 9, 2004

20

Rate Estimation TFRC AlgorithmRate Estimation TFRC Algorithm

On receiving an ack:1. Compute R (the original TCP Friendly rate) .2. Estimate rttopt. using the r(N) approximation.

[Assume N can be obtained from the routing protocol.]

3. Compute the adjusted loss event rate p’ using rttopt and R.

4. Compute the estimated optimum send rate R’.5. Use the original rate, R, if the new rate, R’, is

larger.6. If there is a rate change, make the change

incrementally as TFRC does.

DMS 2004 September 9, 2004

21

OutlineOutline

Introduction Background TFRC Performance over Wireless

Network RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

22

Simulation DetailsSimulation Details NS-2 was used to simulate and evaluate

RE-TFRC performance. Wireless Multi-hop Chain Network

– N-hop network implies N+1 nodes (n0 to nN).

– All simulated TRFC flows go from n0 to nN.– The number of hops in the chain network

was varied from 4 to 15.– The bit err rate (BER) was varied from 10-6

to 10-4.

DMS 2004 September 9, 2004

23

Seven-Hop Chain TopologySeven-Hop Chain Topology

CDF of MAC layer retransmissions

DMS 2004 September 9, 2004

24

Loss Event Rate for Multi-Hop ChainsLoss Event Rate for Multi-Hop Chains

Average loss event rate versus number of hops

DMS 2004 September 9, 2004

25

Average round trip time versus number of hops

Round Trip Times for Multi-Hop ChainsRound Trip Times for Multi-Hop Chains

DMS 2004 September 9, 2004

26

Throughput for Multi-Hop ChainsThroughput for Multi-Hop Chains

Average throughput versus number of hops

DMS 2004 September 9, 2004

27

Average loss event rate for various flow scenarios

Loss Event Rate for Multi-Flow TestsLoss Event Rate for Multi-Flow Tests

DMS 2004 September 9, 2004

28

Average round trip time for various flow scenarios

Round Trip Time for Multi-Flow TestsRound Trip Time for Multi-Flow Tests

DMS 2004 September 9, 2004

29

Throughput for Multi-Flow TestsThroughput for Multi-Flow Tests

Average throughput for various flow scenarios

DMS 2004 September 9, 2004

30

Bit Error Rate Test of RE-TFRCBit Error Rate Test of RE-TFRC

BER 10-6 10-5 10-4

RTT Reduction 39% 32% 14%Loss Rate Reduction 55% 45% 29%

Throughput Improvement 6.5% 4.2% 0.5%

Single flow, seven-hop chain topology

DMS 2004 September 9, 2004

31

OutlineOutline

Introduction Background TFRC Performance over Wireless

Networks RE-TFRC Algorithm Performance Evaluation Conclusions and Future Work

DMS 2004 September 9, 2004

32

ConclusionsConclusions Rate Estimation TFRC (RE-TFRC)

– Estimates MAC layer saturation and controls the TFRC sending rate.

– Lowers the delay and loss rate and can even increase throughput in most cases:

• Lowers round-trip time up to 40%• Lowers loss event rate up to 80%• Increases throughput up to 5%.

– reduces MAC layer congestion.

DMS 2004 September 9, 2004

33

Future WorkFuture Work

Extend Algorithm:– To other topologies: cross, grid, and

random– Consider mobile nodes.

Incorporate into applications– Such as streaming multimedia

Implement TFRC wireless extension in Linux.

Thanks!

[email protected]