Distributed Token Circulation in
Mobile Ad Hoc Networks
Navneet Malpani, Intel Corp.
Nitin Vaidya, Univ. Illinois Urbana-Champaign
Jennifer Welch, Texas A&M Univ.
Presented at Int’l Conf. on Network Protocols, Nov 2001
http://faculty.cs.tamu.edu/welch/papers/icnp01.ps or pdf
Introduction
• Mobile Ad Hoc Networks (MANETs)
– Formed by a collection of wireless mobile hosts,
without making use of any existing infrastructure
(such as base stations or telephone lines).
– Pair of nodes communicate with each other either over
a wireless link between the two nodes, or by traversing
a sequence of wireless links over several other
intermediate nodes.
Example Mobile Ad Hoc Network
A
B
C
D
A
B
D
E
C
E
Introduction continued
• Usefulness
– Disaster recovery
– Search and rescue in remote areas
– Military operations
• Characteristics of Mobile Ad Hoc Networks
– Highly dynamic topology
– Highly variable message delays
– Variable transmission error rates
– Constraints on energy consumption
– Constraints imposed by wireless interfaces
Token Circulation Definition
• Ensure that a token circulates throughout
the network, visiting every node infinitely
often.
• Round: Minimal length sequence of nodes
visited to ensure that every node is visited at
least once.
Token Circulation Example
T
A
B T
A B C E D C A B C E D C A ...
T C T
D
T
E
T
Length of round 1: 5
Length of round 2: 6
Length of round 3: 6
Token Circulation Application
• Total order of message delivery in a group
communication service
• Key features of a group communication
service:
– Maintaining information regarding group
membership
– Communication among nodes in the group in
an ordered manner
Token Circulation Application
• Token carries a sequence number, which is
always incremented. Sender multicasts
message with sequence number; receiver
delivers in order. OR
• Messages are stored in the token itself
(large token).
• Additional mechanisms are needed to obtain
desired level of reliability.
Token Circulation Algorithms
• Local Least Recently Visited (LR): forward
token to neighbor visited least recently
• Local Least Frequently Visited (LF):
forward to neighbor visited least frequently
A
B
LR: ACBCDE CACBCDE CACBCD E...
C
D
LF: ACBCDE DECACB CDEDEDECACB C...
E
More TC Algorithms
Choose next destination among all nodes.
• Global Least Recently (GR): forward to any
node in network visited least recently
• Least Frequently (GF): forward to any node
in network visited least frequently
Yet More TC Algorithms
• GRN: Global Least Recently + visit
intermediate nodes on the path
• GFN: Global Least Frequently + visit
intermediate nodes on the path (not studied)
• Iterative Search: try to find Hamiltonian
Path using more history information (see
paper for more details)
Performance Measures
• Round length: number of nodes visited by
the token in a round
• Message overhead: number of bytes sent
per round
• Time overhead: time required to complete a
to complete a round
Simulation Results
• ns-2 simulator with CMU extensions
• System contains 20 nodes initially placed
randomly in a 1000m x 300m box
• Random Waypoint mobility model
• Each algorithm runs as an application on
top of TCP and DSR protocol
• Results for Static and Dynamic topologies
Static Topologies
• Plots of
– number of nodes visited
– number of bytes sent
– amount of time taken
during each round, averaged over 50 different
scenarios
80
70
LF Algorithm
60
LR Algorithm
50
GF Algorithm
GR Algorithm
40
GRN Algorithm
30
20
10
Round Number
76
71
66
61
56
51
46
41
36
31
26
21
16
11
6
0
1
Average Number of Nodes Visited / Round
Iterative Algorithm
100000
Iterative algorithm
90000
LF Algorithm
LR Algorithm
70000
GF Algorithm
60000
GR Algorithm
50000
GRN Algorithm
40000
30000
20000
10000
Round Number
77
73
69
65
61
57
53
49
45
41
37
33
29
25
21
17
13
9
5
0
1
Average Number of Bytes / Round
80000
1.4
Iterative Algorithm
LF Algorithm
LR Algorithm
1
GF Algorithm
0.8
GR Algorithm
0.6
GRN Algorithm
0.4
0.2
Round Number
77
73
69
65
61
57
53
49
45
41
37
33
29
25
21
17
13
9
5
0
1
Average Amount of Time / Round
1.2
Discussion of Static Results
• LF diverges
• GR and GF trivially have best round length,
but not so good on messages & time
• LR is quite good
• Iterative Search is best overall
Dynamic Topologies
• Varying speed (6, 12, 18 and 24 m/sec) with
constant hello interval of 0.5 sec
• Varying hello interval (0.1, 0.3, 0.5 and 0.7
sec) with constant speed of 12 m/sec
• Hello Threshold: 3
• Number of scenarios: 30
• Duration of simulation was varied inversely
with the speed
35
Average Number of Nodes Visited / Round
Iterative - no hello
30
Iterative - with hello
LF
25
LR
20
GF
GR
15
GRN
10
5
0
6
12
18
Speed (m/sec)
24
60000
Average Number of Bytes / Round
Iterative - no hello
Iterative - with hello
50000
LF
40000
LR
GF
30000
GR
GRN
20000
10000
0
6
12
18
Speed (m/sec)
24
2
Iterative - no hello
1.8
Average Amount of Time / Round
Iterative - with hello
1.6
LF
1.4
LR
1.2
GF
1
GR
0.8
GRN
0.6
0.4
0.2
0
6
12
18
Speed (m/sec)
24
40
Iterative - no hello
Average Number of Nodes Visited / Round
35
Iterative - with hello
LF
30
LR
25
GF
20
GR
GRN
15
10
5
0
0.1
0.3
0.5
Hello Interval
0.7
60000
Iterative - no hello
Iterative - with hello
Average Number of Bytes / Round
50000
LF
40000
LR
GF
30000
GR
GRN
20000
10000
0
0.1
0.3
0.5
Hello Interval
0.7
2
Iterative - no hello
Iterative - with hello
1.6
Average Amount of Time / Round
LF
LR
1.2
GF
GR
0.8
GRN
0.4
0
0.1
0.3
0.5
Hello Interval
0.7
Discussion of Dynamic Results
• Random Nature of Results
– Effect of uncertainty in the topology knowledge
due to the hello protocol
– Effect of the TCP timeout intervals when
partitions occur
– Chaotic nature of the algorithms themselves
• LR is the best! Close to optimal round
length.
Conclusion
• Identified new problem for MANETs -token circulation
• Proposed several distributed algorithms
• Compared them by simulation
• Overall best algorithm :
– Iterative Search in the static case
– LR algorithm in the dynamic case
Future Work
• Identify characteristics of graphs on which LR has
good performance -- there are graphs on which it
has exponential round length (cf. recent work by
Yu Chen)
• Integrate token circulation with mechanisms for
complete group communication service
• Make tolerant of token loss / partitions
• Find lower bounds on possible performance and
find optimal algorithms