Opportunistic data collection through delegation

We consider a collection system where collectors move around gathering information generated by data producers. In such a system, data may remain uncollected when the number of collectors is insufficient to cover the whole population of producers. Motivated by the observation that node encounters are sufficient to build a connected relationship graph, we propose to take advantage of the inherent interactions among nodes and transform some producers into delegates. With such an approach, collectors only need to meet delegates that, in turn, are responsible for gathering data from a subset of standard producers. We achieve this goal through two contributions. First, we investigate several delegation strategies based on the relative importance of nodes in their social interactions (i.e., the node centrality). Second, by considering a prediction strategy that estimates the likelihood of two nodes meeting each other, we investigate how the delegation strategies perform on predicted traces. We evaluate the delegation strategies both in terms of coverage and size of the delegation existing real mobility data sets. We observe that delegation strategies that rely on localized information perform as well as the ones that consider a complete view of the topology. Key-words: collaborative data collection; centrality; sparse networks; social interactions in ria -0 05 08 27 3, v er si on 2 9 M ar 2 01 1 Opportunistic data collection through delegation Résumé : Nous considérons un système de collecte où les collectionneurs se déplacent et collectent les informations générées par les producteurs de données. Dans un tel système, les données peuvent ne pas être collectées lorsque le nombre de collectionneurs est insuffisant pour couvrir l’ensemble de la population des producteurs. Motivé par le fait que les rencontres de nIJuds sont suffisants pour construire un graphe connecté, nous proposons de profiter des interactions inhérentes entre les nIJuds et transformer certains producteurs en délégués. Avec une telle approche, les collectionneurs ont seulement besoin de rencontrer les délégués que, à leur tour, sont responsables de la collecte de données d’un sous-ensemble des producteurs. Nous atteignons cet objectif grâce à deux contributions. Tout d’abord, nous étudions plusieurs stratégies de délégation basée sur l’importance relative des nIJuds dans leurs interactions sociales (par exemple, la centralité du nIJud). Deuxièmement, en considérant une stratégie de prédiction qui donne les estimations de la probabilité d’une rencontre de deux nIJuds, nous étudions les stratégies de délégation avec les traces prédit. Nous évaluons les stratégies de délégation à la fois en termes de couverture et de la taille du groupe de délégation en utilisant des traces de mobilité réelles. Nous n’observons que les stratégies de délégation qui se basent sur des informations localisées fournis aussi des bons résultats comparés aux résultats considérant une vue complète de la topologie. Mots-clés : collecte collaborative de données, centralité, les réseaux de faible densité, les interactions sociales in ria -0 05 08 27 3, v er si on 2 9 M ar 2 01 1 Opportunistic data collection through delegation 3 1 Context and motivation Cellular phones have recently been considered as a pervasive mobile sensing platform due to their increasing proliferation and multiple advanced capabilities (e.g., cameras, GPS, sensors, wireless communication). As a result, global sensing has appeared as one of the most challenging pervasive applications aiming at improving the quality of life of the population [21]. In such applications, also referred to as participatory sensing, mobile nodes (or producers) generate data from observations of the surrounding environment or from users (healthy, activity, behavior,etc) and send it to a central entity that evaluates the global behavior of the system from localized views. In addition, the underlying wireless network may face problems of intermittent and/or sparse connectivity, high degree of mobility, and unreliable links, which greatly impairs the effectiveness of both data collection and delivery. There are many possibilities of how sensed data can be sent to the central entity. When possible, the most straightforward one is to rely on a reliable deployed infrastructure that maintains permanent connectivity with each one of the nodes, e.g., the 3G network. Nevertheless, there are many situations where the use of such infrastructure is prohibitive, either because of cost constraints or capacity limitations. Indeed, depending on the target scenario, information generated by each node might be huge, requiring a significant amount of communication resources to transfer sensed data to the infrastructure (e.g, multimedia contents). We propose to design a collection system composed of specialized devices called collectors whose role is to move around and collect data when they enter within communication range with producers. As part of a general-interest system, collectors can be provided by administrative entities in order to alleviate end-users from using a costly deployed infrastructure (e.g. 3G access). Our idea is to use direct communications among nodes whenever available and use that infrastructure as little as possible. There are two main issues that arise when designing such a system: (i) the number of producers might be much larger than the number of collectors and (ii) we need to decide which collector to “visit” which producer, a challenging problem considering the previous issue. To address these challenges, we need alternative solutions to overcome the insufficiency of the collection system. We suggest to benefit from the inherent contact opportunities that emerge due to the natural mobility of the nodes to tackle these problems. To the best of our knowledge, this is the first time that this specific problem is addressed. Indeed, although previous work makes use of mobile entities to perform some network task, the perspective therein is different from ours [11, 14, 25]. In particular, data mules and message ferries are mobility-assisted strategies that aim at providing “connectivity opportunities” among nodes [14, 25]. These approaches are concerned with controlling the trajectories of ferries or mules to meet static or mobile nodes with the goal of minimizing message drops. Trajectories are adjusted based on requests from nodes, or the nodes adjust themselves their trajectories to meet the ferries. These approaches only partially solve the problem considered in this paper. Indeed, we do not deal with collectors placement or their trajectory design to visit all producers in the network. We instead focus on detecting which subset of producers could help collectors in the data collection task, while keeping their “normal” behavior within the network. More specifically, we propose to take advantage of the inherent mobility of producers and transform some of them into delegates of the collection system. Delegates are responsible to collect data from other producers and serve as intermediate relays between them and the collectors (see Fig. 1). This is motivated by the fact that social RR n° 7361 in ria -0 05 08 27 3, v er si on 2 9 M ar 2 01 1 4 Bigwood & Carneiro Viana & Boc & Dias de Amorim