The
Institute for
Systems
Research
Component Based Models for Performance Analysis and
Design of Ad-Hoc Routing Protocols
HyNet
Kiran K. Somasundaram, Kaustubh Jain, Vahid Tabatabaee, John S. Baras
Objective and Approach:
Selector of Topology Information (MPR)
to Disseminate (STIDC):
• A formal method for design and analysis of ad-hoc routing protocols.
• Specification of the main components of a routing protocol
• Specification of the performance metrics for each component.
• The main components are:
•  Neighborhood Discovery Component (NDC)
•  Selector of Topology Information to Disseminate
•  Route Selection
• Design methodology for each component
• Every node selects a subset of its neighbors as its MPR nodes.
• Only links from selected MPR to the MPR seector are advertised.
• MPR nodes are selected to cover all second order neighbors of a node.
• MPR nodes will also be used for information broadcasting in network
Proposed Design Methodology:
• The design objective is to design a network with predictable and robust performance
• NDC parameters should be selected to control timeliness of link detection.
• Low detection time for good links
• Short lifetime for bad links
Implemented Fixed Point model:
Mapping 2
Selector topology
Info. Dissemination
MPR Probability
Mapping 1
Neighborhood discovery
Link Probability
Mapping 3
Route Selection
Next hop Probability
Mapping 5
MAC & PHY layer models
Packet Failure and
Service Time
• The STIDC objective is to select MPR from stable nodes to control the overhead of
information broadcasting and route selection in the network.
Let
denote the rate of the bidirectional ON-OFF process observed at host “h” for
every neighbor “j” in N1(h).
Mapping 4
Packet forwarding
Scheduling Rate
such that shortest paths from “h” to 2-hop neighbors are preserved.
Neighborhood Discovery:
• We show that the problem is NP-Hard. Greedy Approximation Algorithm is used.
Expected Link Detection Delay vs. U for 0.8 Hello Message Success Probability
3
Expected Detection Delay in 1/!H
10
• Periodic HELLO messages to detect bidirectional links.
• Modelled as a finite state Markov Chain.
• Control parameters are U and D.
• Input is transmission success probability
• Performance metrics are computed: (1) Link detection probability, (2)
Delay in detection of a link, (3) Life time of a link
2
10
We set U=2 and D=2 to control
delay of the NDC algorithm
1
10
0
10
1
2
U-1
0
4
10
U
2
4
6 Link Life Time
8
12
14Probability 16
Expected
vs. D for10
0.6 Hello Message
Success
U
18
U+D-1
U+2
U+1
Example Network
Good link detection delay
Communication
overhead of our
algorithm vs MPR
Heuristic
3
Finite State
Machine for
the NDC
Expected Life Time in 1/!H
10
U+D
20
2
10
1
10
0
10
1
2
3
4
5
6
D
Bad link life time
7
8
9
10