Final Report Summary - RETUNE (Resilience of Opportunistic Networks to Node Misbehaviors)
The primary aim of this grant's research has been the assessment of the resilience of opportunistic networks to node misbehaviours and the proposal of countermeasures / ways to cope with them. Since opportunistic networks constitute an instance of user-centric networks, whose existence and maintenance per se depend on highly autonomous and independent decisions of end-user nodes as to whether they will contribute their resources, it is deemed mandatory to study all those factors that shape their behaviour and determine whether they will cooperate with other user-nodes towards realising the opportunistic network or not. Notably, these factors bear multiple dimensions, such as economical, social, and psychological, making the research highly interdisciplinary and broadening the range of research methodologies and tools.
Overall, a great part of the project research work was devoted to the assessment and better understanding of the impact of node misbehaviours / imperfect cooperation on the performance of opportunistic networking. Initial research work, which also served as motivation for this project, focused on fundamental expressions of node misbehaviours (deferral from packet copying and packet forwarding) and provided analytical models for the (degraded) performance of known protocols (epidemic, two hop) in their presence. Common modelling approach to this work is the formulation of the protocols' operations under imperfect cooperation into absorbing continuous time Markovian chains (CTMCs). What changes from protocol to protocol are the transition rates between transient states and towards the absorbing states. Despite its neatness and novelty, this approach is only applicable to scenarios of uniformly random nodes' movements, i.e. scenarios that can support the exponentially distributed inter-contact times assumption and the CTMC formulation.
To cope with this, we proposed an innovative method for the performance analysis of a broad range of opportunistic forwarding protocols over logs of encounters between mobile nodes (contact traces). The method is applied at the message level, is modular and evolves in three main processing steps. First, the original trace is filtered to isolate those contacts that constitute message forwarding opportunities under the rules of the specific protocol. These forwarding contacts are then captured into sparse space-time graph representations (constructs) with the help graph expansion techniques. Finally, standard shortest path algorithms are run over these constructs to derive typical performance metrics such as message delivery probability and delay, path hopcount, and generated message overhead. Moreover, the original traces can be optionally inflated to improve the accuracy of the method at the expense of additional processing overhead. The method was originally applied to the analysis of opportunistic protocols under a broader range of misbehaviours (including 'social selfishness' expressions, black hole attacks etc) but it soon became apparent that it can be leveraged to a more general performance analysis tool that can used for the study of various unicast forwarding schemes, both of the controlled-flooding type and utility-based, and implementations of multicast forwarding. Comparing the method against discrete event simulations, we found excellent match with the simulation results at run times up to three orders of size lower than those of the simulations.
In parallel with his work on performance analysis, the project work has also been directed towards the impact of imperfect cooperation on one of the fundamental, yet not adequately explored, primitives of all multi-copy message forwarding schemes, the message replication strategy. In contrast with the relaying of message copies, whereby all protocols rely to some extent on the assistance of intermediate nodes, the generation of new message copies may either be exclusively source-controlled or more aggressively involve intermediate relay nodes. Whereas the delegation of the message replication task to intermediate nodes results in faster availability of the message replicas in the network under nominal full node cooperation conditions, it is not necessarily so when full cooperation is relaxed. We first formulated and solved analytical models for the performance of the two message replication strategies under imperfect node cooperation and obtained numerical results:
(a) showing that source replication is consistently more robust to imperfect cooperation than binary replication, i.e. the optimal variant of intermediate node-assisted replication; and
(b) establishing the degree of cooperation that indeed reverses their performance ordering.
The analytical conclusions have been supplemented and verified by trace-driven investigations and eventually, motivated the introduction and analysis of an alternative message replication function that better resolves the 'aggressive replication' versus 'securing replicas' availability' trade off inherently present in the two aforementioned strategies.
The other major thread of the project research work has looked at the causes that can motivate and shape the cooperative / competitive behaviours and actions of rationally selfish (as opposed to irrationally malicious) nodes. These causes are closely related to the particular service / application dynamics but also embody more persistent economical, social and psychological factors. The relevant research work has centered on a particular service paradigm, namely opportunistic parking assistance systems in smart (a.k.a. networked and sensor-equipped) urban environments with two types of parking resource facilities: the public on-street parking (private parking lot) space is assumed to be limited (unlimited) and less (more) expensive, while an additional cruising cost is incurred when resorting to a private parking lot after a failure to find a public parking spot. Our research has tried to analyse how the different factors affect the user nodes' behaviours and determine the efficiency of these systems. Methodologically, we have approached user nodes as selfish non-cooperative agents that strategically interact in an attempt to maximise their own benefit. We have first considered the ideal reference scenario, where user-nodes are fully rational agents and possess perfect information that lets them iterate over all possible outcomes of their actions and choose the one that gives rise to a Nash equilibrium. Then, we have step-by-step relaxed this assumption by accommodating in our game formulations several expressions of bounded rationality such as:
a) the amount of information the opportunistic system can provide the user nodes (drivers) with; we have considered both probabilistic and uncertainty scenarios deriving the respective Bayesian / pre-Bayesian games;
b) psychological biases and cognitive heuristics that are usually employed in human decision making.
In each case, we have derived the respective equilibria, which rely on different solution concepts such as Bayesian Nash, cumulative prospect theory, and quantal response equilibrium. The equilibrium assignments of parking space are compared against the optimal assignment that could be determined by an ideal centralised parking spot reservation system and their price of anarchy is computed. The results of this work have let us draw practical hints as to how application parameters (such as pricing and location of parking facilities) and the information availability do shape the competition level that emerges in the network and generate different equilibria states corresponding to different levels of service efficiency. Notably, the results have pointed to counterintuitive less-is-more effects about the way information availability modulates the service efficiency, promoting the role that information management can play in user-centric networks with nodes competing for some resources.
Finally, we have made the first steps towards a performability-oriented assessment of opportunistic protocols, where the notions of performance and reliability are jointly considered under a single fitness index. This work can effectively bridge mature work on system performance and reliability (performability) with opportunistic networking challenges and will see further work after the end of the RETUNE project.
Overall, the RETUNE project work essentially motivates a reiteration of the basic principles of forwarding protocol design and evaluation in opportunistic networks. In the same time, it is, in its entirety, an argument in favour of resilience as a major design primitive for opportunistic forwarding. By closing this gap in the way opportunistic forwarding is viewed and assessed, it makes a modest contribution so that the opportunistic networking paradigm does skip the fate of other much promising paradigms (e.g. mobile adhoc networks) and find a path towards its broad commercial use in real world as an alternative or complementary component to more interactive forms of networked communication.
Overall, a great part of the project research work was devoted to the assessment and better understanding of the impact of node misbehaviours / imperfect cooperation on the performance of opportunistic networking. Initial research work, which also served as motivation for this project, focused on fundamental expressions of node misbehaviours (deferral from packet copying and packet forwarding) and provided analytical models for the (degraded) performance of known protocols (epidemic, two hop) in their presence. Common modelling approach to this work is the formulation of the protocols' operations under imperfect cooperation into absorbing continuous time Markovian chains (CTMCs). What changes from protocol to protocol are the transition rates between transient states and towards the absorbing states. Despite its neatness and novelty, this approach is only applicable to scenarios of uniformly random nodes' movements, i.e. scenarios that can support the exponentially distributed inter-contact times assumption and the CTMC formulation.
To cope with this, we proposed an innovative method for the performance analysis of a broad range of opportunistic forwarding protocols over logs of encounters between mobile nodes (contact traces). The method is applied at the message level, is modular and evolves in three main processing steps. First, the original trace is filtered to isolate those contacts that constitute message forwarding opportunities under the rules of the specific protocol. These forwarding contacts are then captured into sparse space-time graph representations (constructs) with the help graph expansion techniques. Finally, standard shortest path algorithms are run over these constructs to derive typical performance metrics such as message delivery probability and delay, path hopcount, and generated message overhead. Moreover, the original traces can be optionally inflated to improve the accuracy of the method at the expense of additional processing overhead. The method was originally applied to the analysis of opportunistic protocols under a broader range of misbehaviours (including 'social selfishness' expressions, black hole attacks etc) but it soon became apparent that it can be leveraged to a more general performance analysis tool that can used for the study of various unicast forwarding schemes, both of the controlled-flooding type and utility-based, and implementations of multicast forwarding. Comparing the method against discrete event simulations, we found excellent match with the simulation results at run times up to three orders of size lower than those of the simulations.
In parallel with his work on performance analysis, the project work has also been directed towards the impact of imperfect cooperation on one of the fundamental, yet not adequately explored, primitives of all multi-copy message forwarding schemes, the message replication strategy. In contrast with the relaying of message copies, whereby all protocols rely to some extent on the assistance of intermediate nodes, the generation of new message copies may either be exclusively source-controlled or more aggressively involve intermediate relay nodes. Whereas the delegation of the message replication task to intermediate nodes results in faster availability of the message replicas in the network under nominal full node cooperation conditions, it is not necessarily so when full cooperation is relaxed. We first formulated and solved analytical models for the performance of the two message replication strategies under imperfect node cooperation and obtained numerical results:
(a) showing that source replication is consistently more robust to imperfect cooperation than binary replication, i.e. the optimal variant of intermediate node-assisted replication; and
(b) establishing the degree of cooperation that indeed reverses their performance ordering.
The analytical conclusions have been supplemented and verified by trace-driven investigations and eventually, motivated the introduction and analysis of an alternative message replication function that better resolves the 'aggressive replication' versus 'securing replicas' availability' trade off inherently present in the two aforementioned strategies.
The other major thread of the project research work has looked at the causes that can motivate and shape the cooperative / competitive behaviours and actions of rationally selfish (as opposed to irrationally malicious) nodes. These causes are closely related to the particular service / application dynamics but also embody more persistent economical, social and psychological factors. The relevant research work has centered on a particular service paradigm, namely opportunistic parking assistance systems in smart (a.k.a. networked and sensor-equipped) urban environments with two types of parking resource facilities: the public on-street parking (private parking lot) space is assumed to be limited (unlimited) and less (more) expensive, while an additional cruising cost is incurred when resorting to a private parking lot after a failure to find a public parking spot. Our research has tried to analyse how the different factors affect the user nodes' behaviours and determine the efficiency of these systems. Methodologically, we have approached user nodes as selfish non-cooperative agents that strategically interact in an attempt to maximise their own benefit. We have first considered the ideal reference scenario, where user-nodes are fully rational agents and possess perfect information that lets them iterate over all possible outcomes of their actions and choose the one that gives rise to a Nash equilibrium. Then, we have step-by-step relaxed this assumption by accommodating in our game formulations several expressions of bounded rationality such as:
a) the amount of information the opportunistic system can provide the user nodes (drivers) with; we have considered both probabilistic and uncertainty scenarios deriving the respective Bayesian / pre-Bayesian games;
b) psychological biases and cognitive heuristics that are usually employed in human decision making.
In each case, we have derived the respective equilibria, which rely on different solution concepts such as Bayesian Nash, cumulative prospect theory, and quantal response equilibrium. The equilibrium assignments of parking space are compared against the optimal assignment that could be determined by an ideal centralised parking spot reservation system and their price of anarchy is computed. The results of this work have let us draw practical hints as to how application parameters (such as pricing and location of parking facilities) and the information availability do shape the competition level that emerges in the network and generate different equilibria states corresponding to different levels of service efficiency. Notably, the results have pointed to counterintuitive less-is-more effects about the way information availability modulates the service efficiency, promoting the role that information management can play in user-centric networks with nodes competing for some resources.
Finally, we have made the first steps towards a performability-oriented assessment of opportunistic protocols, where the notions of performance and reliability are jointly considered under a single fitness index. This work can effectively bridge mature work on system performance and reliability (performability) with opportunistic networking challenges and will see further work after the end of the RETUNE project.
Overall, the RETUNE project work essentially motivates a reiteration of the basic principles of forwarding protocol design and evaluation in opportunistic networks. In the same time, it is, in its entirety, an argument in favour of resilience as a major design primitive for opportunistic forwarding. By closing this gap in the way opportunistic forwarding is viewed and assessed, it makes a modest contribution so that the opportunistic networking paradigm does skip the fate of other much promising paradigms (e.g. mobile adhoc networks) and find a path towards its broad commercial use in real world as an alternative or complementary component to more interactive forms of networked communication.