Cooperative and Reliable Packet-Forwarding on Top of AODVBracha Hod March 2006

Outline

Background

Mobile ad hoc network Ad-hoc On Demand Distance Vector Trust and reputation

Problem statement Solution

Misbehaving detection Reputation system Misbehavior reaction

Simulation results Conclusions

Mobile Ad hoc Network

An autonomous, self-configuring system of mobile devices (laptops, smart phones, sensors, etc.) connected by wireless links Each node operates as both an end-system and a router MANET characteristics:

Mobility and dynamic topology Bandwidth-constrained Energy-constrained Prone to security threats

Mobile Ad hoc Network

MANET Routing Protocols

Proactive/Table-driven

Periodically broadcast information across the network in a controlled flood

Waste bandwidth and power consumptionInitiate a route only when it is required Delay when building new routes

Reactive/On-demand

Ad-hoc On-demand Distance Vector

RFC 3561 (2003)One of the leading protocols for MANET Uses sequence numbers to avoid loops Quick adaptation to dynamic networks Low processing and memory overhead Scalable

AODV Route DiscoveryRoute Request Reverse Route Route ReplyA B

C

G E F

D

AODV Route MaintenanceHello Message Route Error

A

B

C

G E F

D

Trust and Reputation

Trust

A subjective expectation a node has about another nodes future behavior, based on the history of their encounters A perception that a node creates through past actions about its intentions and norms A system in which the nodes who participate in it compute rating values and then advertise these values among the other nodes

Reputation

Reputation System

Problem Statement

MANET is vulnerable to many attacks

Packet dropping is the most common attackSelfish nodes are interested in saving their battery life Malicious nodes aim to damage other nodes Black hole node advertises itself as part of a path and then drop the packets Gray hole node adversary selectively drops some packets but not other

Motivation to misbehave

Misbehavior patterns we handle

Solution

Misbehavior Detection

Watch the neighbors and record their behavior

Reputation System

Maintain direct rating according to the observations Exchange rating among nodes Incorporate direct and indirect rating Use trust informationClassify nodes Select reliable paths Punish misbehaving nodes

Misbehavior Reaction

First-Hand Observations

Overhear neighbors

Direct mode getting packets explicitly Promiscuous mode

Examine the overheard packets Update the positive and negative actions

k

i j h

Direct Rating

Calculation and management of the rating using the Beta distribution function Direct rating of a node j by its neighbor i

Rating Exchange

Local model as a result of MANET constrains

Reputation distribution is performed continuously Neighbors direct rating and a black list of misbehaving nodes are exchanged among 1hop neighbors Limited detection and punishment in large and mobile networks

Trust

Misbehaving nodes might spread false rating informationThe trust estimates the reliability of the reports

Second-Hand Observations

Accept indirect rating DRk,j if the node is trusted or if it passes the deviation test Estimate of the indirect positive and negative actions based on the indirect rating Combine the direct and indirect rating to a total rating

Misbehavior Reaction

Nodes classification

Total rating value with total positive and negative actions Two nodes with the same total rating, but with different history are classified differently Greedy selection of the next hop Path maintenance for partial dropping Second chance when the rating is faded

Path selection

Punishment of misbehaving nodes

Simulation Model

Simulation in GloMoSim Standard parameters of the channel and radio model IEEE 802.11 as the medium access protocol Nodes are places randomly in the area Movement by random waypoint model

Speed range of 5-20 m/s Pause time range of 0-500s

Data packets transmission at constant bit rate (CBR) on routes above 1-hop length

Throughput of Well-behaving Nodes50 Nodes 100 Nodes

15 Sources, 15 Black-holes

20 Sources, 30 Black-holes

Punishment of Misbehaving NodesData Packets Transmitted by Misbehaving Nodes Data Packets for Misbehaving Nodes That were not Transmitted

50 Nodes, 15 Sources, 15 Black-holes

Partial Dropping (Gray holes)Data Packets DroppedDropping percentage of 50% (32% of the total rating) Different Dropping Percentages

50 Nodes, 15 Sources, 15 Gray-holes

Robustness against Advanced LiarsData Packets Received False Positives

50 Nodes, 15 Sources, 10 Black-holes

Scalability over AODVThroughput Data Packets Dropped

500 Nodes, 250 static and the remainder walk on speed of 5-10 m/s. 30 Sources, 50 black holes

Conclusions

A reputation system on top of AODV is effective for both partial and complete dropping The reputation system remained robust against advanced liars, when a majority of the nodes are trustworthy In large and unstable networks, it is better to rely on self-observations because the network conditions have greater effect than the reputation system benefits

Thank you!