Ever more powerful mobile devices are nowadays capable of collectively carrying out reasonably demanding computational tasks without offloading the processing to an edge server or a distant cloud-computing service. In this work, we explore such distributed computing and study how it is affected by the mobility as well as the number of nodes that collaborate. We choose distributed counting of a number of nodes in an enclosed area as application. Such application is useful for estimating attendance at events and measuring occupancy for facility management, as needed for monitoring of crowdedness with respect to safety and evacuation, climate control and comfort. For this application, we are interested in determining the time until all nodes know the correct number of nodes in the space where they reside. Our study shows the effect of mobility on the distributed process. We find that the process of collecting data opportunistically from a closed set of nodes is well described by empirical laws that we derive. We discuss the results and suggest further work needed to understand opportunistic computation and to develop it as a new model of computation among collectives of mobile nodes.

A major challenge in mobile wireless devices for opportunistic networks is to decrease the energy consumption. The decrease should not come at a cost of reduced application throughput (i.e., goodput). This work evaluates the potential performance gains for mobile nodes that adopt duty cycling in an opportunistic context. The paper presents an analytical framework for evaluating goodput and energy consumption of nodes based on a probabilistic estimation of effective contact durations, and it validates this framework on a mobility scenario. The study shows that both goodput and energy consumption depend strongly on the distribution of listening durations, and that goodput is independent of the contact rate among nodes. This work also includes extensive trace-driven simulations and demonstrates that duty cycling considerably improves the performance of opportunistic networks by decreasing the energy consumption without significantly affecting the goodput.

Opportunistic networking (a.k.a. device-to-device communication) is considered a feasible means for offloading mobile data traffic. Since mobile nodes are battery-powered, opportunistic networks must be expected to satisfy the user demand without greatly affecting battery lifetime. To address this requirement, this work introduces progressive selfishness, an adaptive and scalable energy-aware algorithm for opportunistic networks used in the context of mobile data offloading. The paper evaluates the performance of progressive selfishness in terms of both application throughput and energy consumption via extensive trace-driven simulations of realistic pedestrian behavior, and demonstrates that the proposed algorithm is robust against the distributions of node density and initial content availability. The results show that in certain scenarios progressive selfishness achieves up to 85% energy savings during opportunistic downloads while sacrificing less than 1% in application throughput. Furthermore, the study demonstrates that in terms of total energy consumption (by both cellular and opportunistic downloads) in dense environments the performance of progressive selfishness is comparable to downloading contents directly from a mobile network.

Direct device-to-device communication based on the `store-carry-forward' paradigm is considered a feasible means for offloading mobile data traffic. Employing device-to-device communication however should satisfy user demand without greatly affecting battery lifetime. In order to provide satisfactory application throughput it is often assumed that nodes behave altruistic, and are willing to share contents for an infinite amount of time. In the context of energy savings, this assumption is overly optimistic. This work evaluates the performance of a publish/subscribe opportunistic content dissemination application that uses duty cycling for energy saving and allows nodes to behave in a selfish manner. Two types of selfishness are introduced: strict and mild. The paper presents the impact of selfishness on both application throughput and energy consumption via extensive trace-driven simulations, and demonstrates that introducing strictly selfish behavior on top of duty cycling leads to great decrease in energy consumption (up to 90% in certain cases) without causing significant loss in application throughput. Moreover, when the duration of the mild selfishness interval is chosen appropriately, mild selfishness can lead to even further decrease in energy consumption while at the same time increasing the application throughput.

Many mobility models currently used for evaluating wireless communication systems have weak resemblance to reality and there is a lack of understanding on which characteristics of human mobility affect system performance. In particular, most current mobility models assume a free flow of nodes and do not consider how mobility is affected by interactions with other persons and with the physical environment. They also assume a closed system, not considering the effect of node arrival and departure. The structure of space in which the mobility occurs is either not considered at all, or only in a limited way. In this work, we address human pedestrian mobility for evaluation of wireless communication to determine which of the aforementioned aspects need to be captured and to what level of detail. We focus on opportunistic communication in the form of ad-hoc and delay-tolerant networks. For the evaluation, we use mobility models from the field of transportation and urban planning that are used for designing and dimensioning public spaces for comfort and safety of pedestrians in rush hour and emergency evacuation. The models capture micro-mobility of pedestrians better than most mobility models used in mobile networking since the application domain requires realistic representation of node interactions with the physical environment and with other nodes. Our results show that the free flow assumption used in most models does not have a significant performance impact. We also conclude that performance is not very sensitive to accurate estimation of the probability distributions of mobility parameters such as speed and arrival process. Our results, however, suggest that it is important to capture the scenario and space in which mobility occurs since these may affect performance significantly.

A major challenge for wireless networks is to minimize the energy consumption in the mobile devices. This work evaluates potential performance gains of a dual-radio architecture in which a low power radio is used to wake up the primary high power radio. Our targeted domain is opportunistic communication directly between devices. We show that it can significantly reduce the energy consumption at a price of slightly decreased goodput. We also examine the effect of the MAC protocol on the performance of the dual-radio system and we point out that in dense scenarios MAC layer protocols, such as of 802.15.4, do not perform well. We observe that information on device density of an environment can be used to address this shortcoming.

In this paper we examine content distribution in opportunistic networks. We design and evaluate strategies by which a node fetches and shares contents on behalf of other nodes, even when the contents are not of its own interest. We propose four relaying/caching options for improving the use of network resources: relay request on demand, hop-limit, greedy relay request, and weighted relay request with allocated caching positions. We also compare our strategies with a strategy from the literature. The proposed strategies are implemented in the OMNeT++ simulator, and evaluated on mobility traces with different churn levels. The results show that the performance of opportunistic caching is strongly dependent on the level of churn in the network. However, we find a strategy that outperforms the rest in all of the explored network settings.

We present a demonstration scenario to validate a middleware architecture that we designed and implemented for distributing content between mobile users in an opportunistic manner. The peer-to-peer networking architecture allows content dissemination between mobile devices without relying on any infrastructure support. We developed a mobile application utilizing the services provided by the implemented middleware. The application facilitates opportunistic content distribution in both one-to-one and one-to-many dissemination modes.