Resilience and opportunistic forwarding: Beyond average value analysis
Introduction
In opportunistic networks, nodes store, carry, and forward messages when they encounter other nodes using short-range wireless communication. This store-carry-forward (SCF) transport service enables the data flow in the network despite the absence of simultaneous end-to-end connectivity. Yet, the network is a system with highly distributed operation, relying on the good will and cooperation of highly heterogeneous, and not always software and hardware-compatible, user nodes. Moreover, the absence of central coordination and control makes it an easier target for malicious attacks.
Inherent resilience against these challenges to the network operation is provided by data replication. Ideally, data travel in the network over diverse space–time paths, involving disjoint physical spaces and different network nodes. In practice, however, the actual data transfer diversity is highly dependent on the mobility patterns of nodes and the rules of the particular forwarding protocol. In general, forwarding protocols prioritize different performance characteristics such as message delivery ratio or buffer usage, and assign different importance to individual nodes during the data transfer. This, in turn, may render them more vulnerable to a particular type of challenge and more resilient to another.
In general, the performance degradation of opportunistic forwarding in the presence of challenges has been dealt with in literature both analytically [1], [2], [3] and with simulations [4], [5], [6]. Common to all these works is that the opportunistic forwarding performance in the presence of a challenge is assessed through averages values of the performance metrics, usually computed over several simulation runs.
On the contrary, in this paper, we compute and plot metric envelopes, whose upper and lower bounds reflect the best- and worst-case response of a metric, e.g., message delivery ratio, to different realizations of a challenge. The motivating remark is that a simple challenge, such as “K selfish nodes” or “M jamming devices” can have a widely different impact on the performance of the opportunistic forwarding, depending on which K nodes behave selfishly or where the M jammers will be physically placed. The metric envelopes implicitly account for the heterogeneity of the opportunistic network nodes in terms of device capabilities and mobility patterns, as well as the varying intelligence of attackers. At the same time, they provide insights that single average values do not. The breadth of the envelope is an indication of how predictably a protocol will perform in the presence of a given challenge; or, equivalently, how much risk is involved in using the protocol in this case. Hence, a protocol with tight metric envelopes may be occasionally preferable to another with better average scores but higher spread of values.
Drawing on earlier work in [7] we use metric envelopes to assess the resilience of three popular forwarding protocols to three representative types of challenges: occasional software/hardware failures, e.g., due to incompatibility of the software/hardware the encountered devices may use; intentional jamming, a typical example of malicious behavior; and free-riding, is a classical instance of non-cooperative behavior emerging in networked settings lacking central coordination and control functionality. The exact computation of the metric envelope values for these challenges would require enumerating all possible challenge realizations, e.g., combinations of K selfish nodes or placements of the M jammer nodes in the physical space. Clearly such an enumeration becomes computationally intractable already for moderate and even small values of K and M. Therefore, we propose heuristics (cues) for inferring “best”- and “worst”-case scenarios for each challenge and approximating the respective metric envelopes. The derivation of best- and worst-case partitioning of nodes into software/hardware compatible groups are formulated as instances of the community detection and weighted coloring problems, respectively; jammers are placed in the areas that rank highest (resp. lowest) with respect to the density of encounters; and free riders are let coincide with the most (least) central nodes with respect to message delivery.
We demonstrate the use of envelope metrics and the additional information they can deliver through simulation scenarios with various synthetic and experimental mobility traces. The envelopes can provide arguments in favor of one protocol over the other when they are indistinguishable with respect to average performance values. Their width provides an indication of how much performance differentiation is possible in the presence of a given challenge and given node mobility patterns and how well random simulation runs may fail in predicting the impact of a challenge.
In summary, the contributions of this paper are highly methodological and include: (i) the formulation and promotion of the metric envelope concept as a tool for assessing the resilience of opportunistic forwarding schemes in a way that explicitly accounts for the node heterogeneity (device capabilities, mobility), and when relevant, attacker’s intelligence (Section 2); (ii) the proposal of heuristics for approximating the worst- and best-case scenarios for representative challenges (Section 3); and (iii) the demonstration of the methodology in the assessment of three popular forwarding protocols under different challenges and mobility patterns (Section 4). We position this work within the broader literature on opportunistic network resilience in 5 and discuss research directions out of it in Section 6.
Section snippets
Assessing resilience: envelopes instead of average values
To assess the performance of forwarding protocols, we consider two standard performance metrics, the message delivery ratio and delay. The message delivery ratio equals the fraction of messages that reach their destinations out of those generated at their sources (ignoring replicas). For every delivered message, message delay equals the time elapsed between the message generation epoch and its arrival at the destination node.
However, and contrary to earlier studies in literature, we are
Heuristics for performance envelope derivation
We consider three representative challenge instances: software/hardware failures and incompatibilities, jamming, and free-riding phenomena. Note that there is no systematic way to a priori find their actual extreme realizations and their exhaustive enumeration is computationally expensive. Therefore, we resort to challenge-specific heuristics in order to infer best- and worst-case realizations that approximate the envelope bounds for each challenge parametrization.
Methodology: tools, traces, protocols
To demonstrate the use of the metric envelope approach, we carry out experiments with two performance analysis tools: the ONE simulator [19] and the trace-parsing Space–Time-Graph tool described in [20]. ONE provides a broad variety of mobility model and forwarding protocol implementations and is the de-facto simulator for opportunistic networks. The Space–Time-Graph tool, on the other hand, parses traces of encounters and yields significantly faster run times for the narrower selection of
Related work
In general, the performance degradation of opportunistic forwarding schemes in the presence of challenges to their operation has been explored both analytically and with simulations. Most studies have looked into the impact of the free-riding phenomenon, whereby nodes do not (occasionally) contribute to the relaying of other nodes’ messages. In [5] Panagakis et al. perform simulations with synthetic random mobility models and let nodes probabilistically abstain from copying and forwarding
Conclusions
We assess the resilience of opportunistic forwarding schemes through the metric envelope concept, which explicitly accounts for the differentiated challenge impact due to node heterogeneity in terms of device capabilities, mobility and, when relevant, attacker’s intelligence. The envelope approach gives a more complete assessment of a protocol’s resilience than average value analysis. As a general rule, the envelope tends to be narrower for traces that have inherently high degree of diversity
Acknowledgements
This research was funded by the ResumeNet project under the EU grant FP7-224619. The work of the second author has been supported by the European Commission under the Network of Excellence in Internet Science project EINS (FP7-ICT-288021). Our paper benefited from the community detection functionality of the radatools v3.2 software, which is provided freely to the research community by Prof. Sergio Gomez and his colleagues.
References (30)
- et al.
Weighted graphs and university course timetabling
Comput. Oper. Res.
(1992) Assessing the vulnerability of DTN data relaying schemes to node selfishness
Commun. Lett. IEEE
(2009)- et al.
Routing in socially selfish delay tolerant networks
Proc. INFOCOM
(2010) - et al.
A framework for routing performance analysis in delay tolerant networks with application to noncooperative networks
IEEE Trans. Parallel Distrib. Syst.
(2012) - A. Keranen, M. Pitkanen, M. Vuori, J. Ott, Effect of non-cooperative nodes in mobile dtns, WoWMoM,...
- A. Panagakis, A. Vaios, I. Stavrakakis, On the effects of cooperation in DTNs, in: Proc. COMSWARE, pp. 1–6,...
- J. Burgess, G.D. Bissias, M.D. Corner, B.N. Levine, Surviving attacks on disruption-tolerant networks without...
- F. Bjurefors, M. Karaliopoulos, C. Rohner, P. Smith, G. Theodoropoulos, P. Gunningberg, Resilience and opportunistic...
Mobile Cellular Telecommunications Systems
(1989)On evaluating the performability of degradable computing systems
IEEE Trans. Comput.
(1980)