© University of St Andrews, UK1 Chapter 14: Incentive-aware opportunistic network routing Greg Bigwood and Tristan Henderson University of St Andrews Routing in Opportunistic Networks
© University of St Andrews, UK2 Problem: Opportunistic networking relies on cooperation between nodes to perform efficiently Opportunistic routing protocols depend on nodes forwarding messages Otherwise nodes must delivery directly to recipient Cooperative forwarding incurs a cost to forwarding nodes Energy Storage Self-interested nodes avoid forwarding cost: Refuse to pass messages on for other nodes
© University of St Andrews, UK3 Outline We discuss attack on opportunistic routing With a focus on selfishness We discuss incentive mechanisms for opportunistic routing Conclude with discussions of outstanding challenges in the area
© University of St Andrews, UK4 Opportunistic Network Routing Frequent disconnections Mobile nodes coming into and out of range Non-static forwarding paths A varied set of nodes in range over time No predictable interaction schedule Nodes are most likely carried by users with diverse and variable mobility patterns Nodes must opportunistically use any available nodes for forwarding
© University of St Andrews, UK5 Cooperation Opportunistic networking necessarily relies on cooperation to perform efficiently If all nodes participate, we find the shortest paths If nodes do not participate in forwarding, we must pass message directly to destination High latency Low delivery ratio Cooperative forwarding involves cost to intermediaries: Storage of ferried messages Energy cost of forwarding
© University of St Andrews, UK6 Selfishness Selfishness: refusing to forward other nodes messages Reduces cost for intermediary Still expect their own messages to be forwarded by others Harms performance of network
© University of St Andrews, UK7 Reality Mining Selfishness Simulation As proportion of selfishness nodes increases, Delivery ratio decreases. Selfishness harms the network.
© University of St Andrews, UK8 Attacks on Opportunistic Routing Manipulation of routes Nodes may alter the delivery path Selective maliciousness Nodes may be malicious only under certain circumstances Selfishness Nodes messages may not reach destination Users’ economically rational desire to preserve battery affects their selfishness
© University of St Andrews, UK9 Selfishness Opportunistic routing protocols, in particular Epidemic routing and Spray-and-wait routing are vulnerable to selfishness (Panagakis et al). Once 30% of the nodes in the network are selfish, performance degrades (Keränen et al). Is there an acceptable amount of selfishness? What if selfish nodes only forward to the destination, but not other intermediaries? Is this acceptable?
© University of St Andrews, UK10 Incentivising Routing Participation Many approaches in traditional networks Bartering Swap messages 1-for-1 Currency Purchase credits to give to other nodes in return for their forwarding service Asynchronous bilateral trading Nodes perform actions that benefit each other, but not necessarily simultaneously Watchdog mechanisms Nodes monitor each others communication to ensure compliance
© University of St Andrews, UK11 Which are appropriate for Opp Nets? Bartering Not all nodes have equal number of messages to exchange Currency No out of band oracle to administer currency Watchdog mechanisms Not many encounters will be observed by a third party Asynchronous bilateral trading Nodes perform actions that benefit each other, but not necessarily simultaneously
© University of St Andrews, UK12 What information do we have? We must rely only on encounters between nodes Nodes can collect opinion data based on their interactions Nodes can use collated opinion data to make decisions about the trustworthiness of individual nodes Encounter tickets Use PKI to generate provable encounter tickets Used to prove messages were exchanges and encounters took place
© University of St Andrews, UK13 How to bootstrap the mechanism? The incentive mechanism must throughout the entire lifetime of the network We need a mechanism to generate initial trust opinion data We can use Self-Reported Social Networks (SRSNs) Use online social network data or similar out of band data to provide information available before network startup These SRSN data may correlate with trustworthyness
© University of St Andrews, UK14 Attack against incentive mechanisms Exploiting friendship mechansisms Do not incentivise nodes to add as many other nodes as “friends”. Increasing trust through epidemic behaviour Malicious nodes may ignore routing protocols to gain credits/currency/inflated ranking Tailgating Generating large numbers of encounter tickets by following nodes Manipulation of control traffic Withholding ranking information Offer non-existent routes
© University of St Andrews, UK15 Attack against incentive mechanisms 2 Defamation Creating false reputation claims to damage other nods Exploiting detection algorithms Exploiting grace periods or allowances made for genuine device limitations such as battery failure Do not encourage nodes to drop old messages (this may be acceptable) Collusion Sybil attacks How do we know a user cannot easily create a new identity
© University of St Andrews, UK16 IRONMAN: Addressing these concerns IRONMAN Incentives and Reputation for Opportunistic routiNg in Mobile and Ad hoc Networks (Bigwood et al) Use SRSN information to bootstrap network Increase personal ranking of nodes considered friends Use encounter histories to detect selfishness No oracles, watchdogs, infrastructure networks nor flooded delivery receipts required
© University of St Andrews, UK17 IRONMAN Detection Mechanism
© University of St Andrews, UK18 Incentive Mechanism Performance Detection Time The time it takes a mechanism to correctly detect selfish behaviour Detection Accuracy The proportion of selfish nodes that were correctly detected as selfish by a mechanism Selfishness Cost The proportion of forwarded messages that were generated as a result of a node creating a message while selfish
© University of St Andrews, UK19 Performance Comparison Simulation of several popular incentive mechanisms Epidemic routing over the Reality Mining Trace We compare the selfishness cost when two proportions of nodes behave selfishly Nodes have finite buffer, energy and message TTL IRONMAN greatly outperforms other mechanisms
© University of St Andrews, UK20 Incentive Summary By bootstrapping the trust mechanism using SRSNS, and using Encounter histories IRONMAN outperforms existing mechanisms IRONMAN suited to particular networking constraints in Opportunistic Networks This demonstrates that Incentive mechanisms designed for opportunistic routing and useful, and motivates future work in this area
© University of St Andrews, UK21 Challenges For Incentive Aware Routing User behaviour Some nodes may behave altruistically except under specific circumstances. Is this acceptable? How can nodes corroborate information? Exact timings difficult in opportunistic network Using social-network information SRSN information has shown to be useful. Can we perhaps classify users based on social network information? Are opportunistic routing patterns similar to social network communication patterns? May lead to cross disciplinary research
© University of St Andrews, UK22 Challenges For Incentive Aware Routing Cross-layer information use Many Opportunistic Routing applications might themselves involve social networks. E.g. crowdsourcing and mobile social networks. Can we use information from the application at the routing layer or (vice versa)? E.g., spammers have their messages dropped? Modeling social network behaviour Advanced simulation Allows for comparison of social networks and network communication networks Predictive user location may improve routing performance
© University of St Andrews, UK23 Challenges For Incentive Aware Routing Academic challenges Collecting datasets is costly A lack of datasets is harming research Datasets are not shared among researches effectively Metrics for analysing incentive mechanism No consensus on how best to compare and analyse the incentive mechanisms for opportunistic networks. What constitutes a fair distribution of forwarding?
© University of St Andrews, UK24 Conclusions Incentive mechanisms will be vital for any opportunistic networking deployment Existing incentive mechanisms from MANETs and DTNs are innapropriate for opportunistic networks Using SRSN information provides incentive mechanisms with a method of bootstrapping their protocols There a many challenges left for opportimostic routing, many of which are cross-discipline problems