icfcc conference real

19
How bad TCP suffers from packet losses and delays over wireless ad hoc network May Zin Oo, Mazliza Othman Faculty of Computer Science & Information Technology University of Malaya Malaysia 3 rd to 5 th April

Upload: um

Post on 09-Jun-2015

775 views

Category:

Technology


0 download

TRANSCRIPT

Page 1: Icfcc conference real

How bad TCP suffers from packet losses and delays over wireless ad hoc network

May Zin Oo, Mazliza Othman

Faculty of Computer Science & Information Technology University of Malaya Malaysia

3rd to 5th April

Page 2: Icfcc conference real

Outlines

‡ What does Ad hoc network mean?

‡ Applying TCP (Transport Layer Protocol) to Ad hoc

network

‡ TCP Variants for Ad hoc network

‡ Ad hoc routing protocols

‡ Simulation Methodology

‡ Experimental results from simulations

‡ Conclusion

Page 3: Icfcc conference real

‡ Self organizing, self healing, distributed

networks

† The wireless devices can communicate

each other without using central access

points in peer-to-peer fashion , but these

are within range of each other.

‡ Decentralized wireless network

† The determination of which nodes

forward data is made dynamically based

on the network connectivity.

AD HOC NETWORK

Page 4: Icfcc conference real

Applying TCP to Ad Hoc Environment

‡ TCP (Transmission Control Protocol)

† To provide flow control, congestion control & avoidance, in-

order and reliable end-to-end delivery of data packet in the

wired networks.

‡ Applying TCP to ad hoc environment,

† TCP has to face many problems such as packet losses due to

congestion and node mobility, wireless high bit errors, and so

on.

‡ Because of these problems, the following TCP versions were

intended for wireless environment:

† Tahoe, Reno, NewReno, Vegas and Westwood

Page 5: Icfcc conference real

TCP Versions for Ad Hoc Network

‡ TCP Tahoe introduced slow start and congestion avoidance.

† Slow start,

† Use at the beginning of a transfer or after timeout

† Start from minimum window size

† Increase window size exponentially

† Congestion avoidance,

† Increase congestion window size by MSS/cwnd bytes for

each ACK received.

† Increase window size linearly

‡ In our simulation, I will analysis the following TCP variants.

† NewReno

† Vegas

† Westwood.

Page 6: Icfcc conference real

NewReno Vegas Westwood

Features

● Slow start● Congestion

Avoidance● Fast Retransmit● Fast Recovery

● Slow Start● Congestion

Avoidance● Rerouting● Persistent

Congestion

● A sender-side modification of the TCP congestion window algorithm

Loss Detectio

n

● Three duplicate ACKs

● Half the congestion window

● Perform Fast retransmit & Fast recovery

● Go slow start if ACK times out

● One duplicate ACK ● Compare timestamp

of ACK to a timeout value

● Retransmit rather than wait for three duplicate ACKs if timestamp is greater than the timeout value

● Three duplicate ACK● End-to-end

bandwidth estimation to discriminate the cause of packet loss

● Use this estimate to compute congestion window and slow start

Comparison of three variants of TCP

Page 7: Icfcc conference real

Ad hoc routing protocols

Table-driven

DSDV

CGSR

WRP

Source-initiated on-demand

AODV

DSR

LMRTORA

ABR

SSR

Ad Hoc Routing Protocols

Page 8: Icfcc conference real

DSDV DSR AODV

● Hop by hop Distance Vector

● Loop freedom

● Per node routing table

Destination Next hop Number of hops Sequence

number

● Source routing● Each packet:

complete route

● Route Discovery● Route Maintenance● Caching

optimizations

● Like DSR On demand route

discovery, maintenance● Like DSDV

hop by hop routing, route sequence numbers

● Route Request● Route Reply

Number of hops to destination

● Maintenance Periodic HELLO

Ad Hoc Routing Protocols

Page 9: Icfcc conference real

‡ Network Simulator (NS-2) was used to study the three variants of TCP over three ad hoc routing protocols.

Ad hoc environment Static and mobile ad hoc networks

Simulation period 540 seconds

Simulation scenario 5 nodes & 15 nodes chain topology

Simulation topology 2300 m x 1000 m

Transmission range of each node 150 m

Bandwidth 11Mbps

Traffic FTP (File Transfer Protocol)

Packet size 1500 bytes

Simulation Methodology

Page 10: Icfcc conference real

0 1 2 3 4

0 1 2 3 4

1 2 3

0 1 2 3 4 5 6 7 14

0 2

2 12

12

6

7

7 8 14

… …… …

Static ad hoc environment ( 5 nodes & 15 nodes )

Mobile ad hoc environment ( 5 nodes & 15 nodes )

Fig 1: Source node 0 connects to destination node 4 in static ad hoc

network

Fig 2: Source node 0 connects to destination node 4 in mobile ad hoc network

Fig 3: Source node 0 connects to destination node 14 in static ad hoc network

Fig 4: Source node 0 connects to destination node 14 in mobile ad hoc network

Simulation Methodology (Continued…)

Page 11: Icfcc conference real

Simulation Results

‡ We simulate each variant of TCP over each routing

protocol in static ad hoc network and mobile ad hoc

network environments and measure how the node

mobility affects TCP and routing protocols in 5-node and

15-node chain topologies.

‡ We have analyzed † Packet loss rates (%) † Data transfer rate† Throughput of received packets at the destination

node.

Page 12: Icfcc conference real

Packet loss rate (%) measurement of DSDV

5 nodes 15 nodes0

0.5

1

1.5

2

2.5

3

3.5

4

New

Ren

o;

0.5

80

00

00

00

00

00

01

New

Ren

o;

3.3

6

Veg

as;

0

.02

00

00

00

00

00

00

01

Veg

as;

0

.89

00

00

00

00

00

00

1

West

wood

; 0

.13

West

wood

; 0

.45

Packet loss rate (%) using DSDV in static ad hoc network

5 nodes 15 nodes0

0.5

1

1.5

2

2.5

3

3.5

4

New

Ren

o;

0.3

4

New

Ren

o;

2.6

5

Veg

as;

0.0

3

Veg

as;

0.0

9

West

wood

; 0

.58

West

wood

; 0

.71

00

00

00

00

00

00

1

Packet loss rate (%) using DSDV in mobile ad hoc network

Figure 1

Figure 2

Page 13: Icfcc conference real

Packet loss rate (%) measurement of AODV

5 nodes 15 nodes0

0.5

1

1.5

2

2.5

3

3.5

4

New

Ren

o;

0.1

1

New

Ren

o;

3.6

8

Veg

as;

0.0

3 Veg

as;

1.6

3

West

wood

; 0

.09

West

wood

; 0

.36

Packet loss rate (%) using AODV in static ad hoc network

5 nodes 15 nodes0

0.5

1

1.5

2

2.5

3

3.5

4

New

Ren

o;

0

New

Ren

o;

3.5

2

Veg

as;

0 Veg

as;

1.0

7

West

wood

; 0

.41 W

est

wood

; 1

.49

Packet loss rate (%) using AODV in mobile ad hoc network

Figure 3

Figure 4

Page 14: Icfcc conference real

Packet loss rate (%) measurement of DSR

5 nodes 15 nodes0

5

10

15

20

25

30

35

New

Ren

o;

30

.26

New

Ren

o;

18

.06

Veg

as;

3.7

5

Veg

as;

2.4

2

West

wood

; 0

.13

West

wood

; 0

.45

Packet loss rate (%) using DSR in static ad hoc network

5 nodes 15 nodes0

5

10

15

20

25

30

New

Ren

o;

27

.69

New

Ren

o;

18

.81Veg

as;

3.6

3

Veg

as;

5.0

6

West

wood

; 0

.96

00

00

00

00

00

00

1

West

wood

; 3

.45

Packet loss rate (%) using DSR in mobile ad hoc network

Figure 5

Figure 6

Page 15: Icfcc conference real

DSDV DSR AODV0

2

4

6

8

10

12

14

Data transfer rate of Vegas over NewReno

5 nodes

15 nodes

Data

tra

nsf

er

rate

(ti

mes)

Comparison of data transfer rate in static environment

‡ Vegas is the fastest data transfer rate under various conditions.

Figure 7

Figure 8

DSDV DSR AODV0

2

4

6

8

10

12

14

Data transfer rate of Vegas over Westwood

5 nodes

15 nodes

Data

tra

nsf

er

rate

(ti

mes)

Page 16: Icfcc conference real

DSDV DSR AODV0

2

4

6

8

10

12

14

Data transfer rate of Vegas over NewReno

5 nodes

15 nodes

Data

tra

nsf

er

rate

(ti

mes)

Comparison of data transfer rate in mobile environment

Figure 9

Figure 10

DSDV DSR AODV0

2

4

6

8

10

12

14

Data transfer rate of Vegas over Westwood

5 nodes15 nodes

Da

ta t

ran

sfe

r ra

te (

tim

es)

Page 17: Icfcc conference real

‡ All TCP variants over DSDV routing protocol can still receive packets

although nodes move out of each other’s transmission range.

‡ This is because DSDV maintains all possible routes to the destination.

‡ Although DSR routing protocol suffer from most packet losses, the

number of received packets at destination nodes are the most.

Measurement of the throughput in mobile ad hoc network

Page 18: Icfcc conference real

Conclusion

Packet loss rate (%)

Static ad hoc network Mobile ad hoc network

5 nodes 15 nodes 5 nodes 15 nodes

DSDV Vegas Westwood Vegas Vegas

AODV Vegas Westwood NewReno & Vegas Vegas

DSR Westwood Westwood Westwood Westwood

In the overview of performance comparison , according to the packet

loss rate (%) ,

‡ NewReno over AODV routing in mobile environment but in less

nodes,

‡ Vegas over DSDV and AODV routing in both environment and

‡ Westwood over DSR routing in both environment are the best and

least packet loss rate.

Page 19: Icfcc conference real