A Human Connectivity Model for Opportunistic Mobile Systems Roberta Calegari, Mirco Musolesi, Franco Raimondi and Cecilia Mascolo Department of Computer Science, University College London Gower Street, London, WC1E 6BT, United Kingdom {r.calegari|m.musolesi|f.raimondi|c.mascolo}@cs.ucl.ac.uk Abstract Opportunistic networking protocols have recently started to emerge in different contexts, ranging from vehicular communications and remote populations connectivity to wildlife monitoring. These protocols are mainly based on the ability to exploit asynchronous communication among hosts who can act as carriers for the messages which are first stored and transported, and then delivered when the destination is reached. At the heart of these protocols is the concept of hosts colocation and connectivity patterns. Often, however, the protocols are evaluated using mobility models which tend not to mirror the connectivity patterns of the domain in which the protocol needs to be applied, failing to give insight into the performance of the protocols in realistic settings. In this paper we propose a different approach: based on the assumption that opportunistic networking protocols are based on colocation (and connectivity), we present a model for connectivity patterns, which can be extracted from real data. To validate our approach, we show how we used the Dartmouth Campus traces as one of the inputs of our framework to generate connectivity traces with a similar be- haviour. 1 Introduction The recent years have seen a growing interest towards op- portunistic networking protocols [7]. The applications of these protocols range from pure delay tolerant network- ing scenarios for the provision of connectivity in presence of intermittent disconnections or network partitions [24], to information dissemination algorithms for specific sce- narios, including vehicular ones [4] and wildlife monitor- ing [16]. At the heart of many of these protocols is the idea that hosts colocation 1 can be exploited to transfer messages from senders to intermediate nodes, such as mobile carri- ers, and then, from carriers to final receivers, possibly with some delay. Therefore, connectivity, more than mobility, is one of the pre-eminent aspects to be considered in the 1 For the purposes of this work, we speak indistinctly of colocation, contact and connectivity, as they are equivalent in our model. design and performance evaluation studies of this class of systems and protocols 2 . Existing mobility models generate random movement traces like the Random Way Point model [11], with no insight into realistic connectivity patterns. However, there is a stringent need of more realistic and sound connectivity patterns for testing mobile systems in the community: for this reason, many research groups have started projects with the aim of collecting traces for different application scenarios, including students patterns in campuses [8, 27], people attending conferences [10] and cities and streets circulation [22]. Repositories have also been created to col- lect all these measurements (e.g., CRAWDAD Project [12] at Dartmouth College). No matter how many traces can be collected, this will al- ways look like a small amount, in many case insufficient, with respect to the variation needed for a thorough analy- sis of the performance of a system. In addition, a sensi- tivity analysis simply cannot be performed using a single set of traces. A number of pioneering works [1, 2, 8, 28] have studied traces in order to gain a better understanding of the real mobility patterns. A key study in this area is the work on connectivity patterns presented by Chaintreau et alii in [6] which illustrates the fundamental insight that contacts duration and inter-contacts time between individ- uals are distributed according to power-law distributions 3 and that these patterns may be used to develop more effi- cient opportunistic protocols. What we have seen until now is that traces are gener- ally used a posteriori to validate the model and not for a priori study of the connectivity properties. For example in a previous work [21], we developed a model based on so- cial networks able to reproduce patterns observable in real traces, especially in terms of colocation duration and inter- 2 Clearly, mobility models generating geographical coordinates are nec- essary to test location-based systems and geo-routing protocols. 3 Power-law distributions are characterised by the following form: P (x)= x -k with k ≥ 0. A power-law distribution is also called scale-free since it remains un- changed to within a multiplicative factor under a re-scaling of the indepen- dent variable x [23]. 1