alleviating the effects of mobility on tcp performance signal strength based link management

Copyright: UC Riverside

Alleviating the effects of mobility on TCP PerformanceSignal Strength based Link ManagementFabius Klemm*, Srikanth Krishnamurthy+

and Satish Tripathi+,* EPFL Lasaunne ,+ University of California, Riverside

Paper Presented by Dr.Nitin Vaidya, UIUC

Presentation at UCLA on June 6th , 2002


Outline

• Motivation for Research

•Using Lower Layer Support to improve TCP performance

• Link Failures – True and False

• Signal Strength based methods to help improve TCP goodput

• Preliminary experiments and results


Motivation

• TCP is unable to differentiate between true and false link failures – former due to mobility, latter due to congestion.

• Implement link layer mechanisms that can help:

• Anticipate real link failures by signal strength measurements – preemptively initiate route discovery.

• Reactively increase power level for transmission upon the detection of a real link failure to salvage TCP packets in transit.

• Requires mechanisms for differentiating between true and false link failures.


Source DestinationLink breaks

RERRRERR

Background -- AODV

•Ad hoc On-Demand Distance-Vector

• Route discovered by queries. RERR message sent upon discovery of a link failure.


RTS

CTS

Revisiting the IEEE 802.11 MAC protocol

•RTS – CTS – DATA – ACK

•Solves the hidden and exposed terminal problem in most cases.


•Neighbor within reach•Mac Layer cannot establish RTS/CTS Handshake•Mac Layer reports link break to upper layers

False Link Failure Reports


How come?

1 2 3 4 5DataRTS

• Transmission Range: 250 m• Interference Range: 550 m

4 is sending Data to 51 is sends an RTS to 2

2 does not send a CTS because it senses the transmission of 4

Node 1 gives up after seven times

False link layer report


Objective

• In this preliminary work, we consider sparse scenarios and use a rather naïve approach to differentiating between true and false link failures.

• In reality, more sophisticated techniques might be needed.


300 x 1500 meters

• Distance between TCP source and sink:– about 1530 m or 8.8 hops in average

Simulation Scenario

•50 mobile nodes + 2 static nodes

•1 TCP connection


0

2

4

6

8

10

12

0 4 8 12 16 20

Maximum speed in m/s

Pa

cke

t lo

ss in

%

• Reduce packet losses due to mobility (correct link breaks)• Reduce packet losses due to false link failure reports

Problems to be solved


Low Mobility: False link failures dominateHigh Mobility: Correct link failures dominate

0%

20%

40%

60%

80%

100%

0 4 8 12 16 20


Mobility-inducedlink breakages

False link failures

Reasons for packet loss


Neighbor ID Timestamp 1 Distance 1 Timestamp 2 Distance 2

3 4.200 200 4.205 201

…

…

• Node computes distance from signal strength – simple model is assumed wherein the attenuation is inversely proportional to the square of the distance.

• The time stamps correspond to the last two instances when the node heard the neighbor.

Each node maintains a Mac layer neighbor table:

Tackling False Link Failures


Tackling False Link Failures

Persistent Mac

–A Node sends RTS packets more than seven times if neighbor is likely to be within transmission range.

– Simple naïve approach.

– Seems to work in the sparse scenarios considered.

– More sophistication may be needed in dense scenarios.


0

2

4

6

8

10

12

0 4 8 12 16 20


Pac

ket l

oss

in %

Unchanged Mac

Persistent Mac

Persistent Mac – Packet Loss


0%

20%

40%

60%

80%

100%

0 4 8 12 16 20


Mobility-inducedlink breakages

False link failures

Persistent Mac – Link Breakages


0

5

10

15

20

25

30

0 4 8 12 16 20


Impr

ovem

ent o

f TC

P

good

put i

n %

Persistent Mac

Persistent Mac - Goodput


0

0.01

0.02

0.03

0.04

0.05

0 4 8 12 16 20


TC

P r

etr

an

smis

sio

ns

pe

r d

eliv

ere

d d

ata

pa

cke

t

Unchanged Mac

Persistent Mac

Persistent Mac – TCP Retransmissions


• Salvage in transit packets:

Source DestinationLink breaks

Lost packets due to Mobility


Salvage Packets

•Two Approaches:

1.Proactive: Predict link breakage and stimulate the source to preemptively initiate a route discovery.

2.Reactive: Re-establish a broken link with a temporary higher transmission power level.


T1 T2current

timefuture time

distance

• Mac layer informs routing layer when next hop is almost out of range

Mac Layer: Proactive

•Nodes use neighbor table to predict node movement in the future:

–Simple prediction: Assume linear node movement


Mac Layer: Reactive

• Node raises transmission power temporarily if it cannot establish an RTS/CTS handshake


1 2

RTS

RTS – Frame contains power value

CTS

Node 2 sends CTS with the same power

Same for Data and ACK

Node 2 moves out of range of Node 1

Reactive Mac


•Node 3 is receiving Data from Node 4

Node 2 does not know about the data transferThe high power CTS collides with the Data at Node 3

•Node 1 establishes a high power link

1 2

RTS

CTS

3

Data 4

Problems!


Source Destination

Stop sending!

Stop sending!

Salvaging Packets

•Routing layer informs source to stop sending

•But: Intermediate nodes keep forwarding packets


Salvaging Packets

•Routing Layer

–Three route states:

•Down: no route

•Up: route ok, answer route requests

•Going Down: “weak route”, use only to salvage packets, do not answer route requests! NEW!


0

2

4

6

8

10

12

0 4 8 12 16 20


Pa

cke

t lo

ss in

%

Unchanged Mac

Proactive LM

Reactive LM

Combined

Results – Packet Loss


0

10

20

30

40

50

60

0 4 8 12 16 20


Imp

rove

me

nt o

f TC

P g

oo

dp

ut

in %

Proactive LM

Reactive LM

Combined

Results - Goodput


0

0.01

0.02

0.03

0.04

0.05

0 4 8 12 16 20


TC

P r

etr

an

smis

sio

ns

pe

r d

eliv

ere

d d

ata

pa

cke

t

Unchanged Mac

Proactive LM

Reactive LM

Combined

Results – Retransmissions per transmitted packet


Conclusions & Future Work

• The methods proposed seem to improve TCP performance by as much as 40 % in the scenarios considered.

• The reactive scheme might cause problems in highly congested scenarios – especially when the network is dense.

• More sophisticated methods may be needed to clearly differentiate between link failures due to mobility and congestion.

• A node might need to more intelligently decide upon when to increase its transmission power level.


Thank You

alleviating the effects of mobility on tcp performance signal strength based link management

Documents