Abstract
We describe a case study application for untethered video dissemination using a hybrid edge-cloud architecture featuring Android devices, possibly organised in WiFi-Direct groups, and Raspberry Pi-based cloudlets, structured in a mesh and also working as access points. The application was tested in the real-world scenario of a Portuguese volleyball league game. During the game, users of the application recorded videos and injected them in the edge-cloud. The cloudlet servers continuously synchronised their cached video contents over the mesh network, allowing users on different locations to share their videos, without resorting to any other network infrastructure. An analysis of the logs gathered during the experiment shows that such portable setups can easily disseminate videos to tens of users through the edge-cloud with low latencies. We observe that the edge-cloud may be naturally resilient to faulty cloudlets or devices, taking advantage of video caching within devices and WiFi-Direct groups, and of device churn to opportunistically disseminate videos.

Abstract
In ubiquitous streaming data sources, such as sensor networks, clustering nodes by the data they produce gives insights on the phenomenon being monitored. However, centralized algorithms force communication and storage requirements to grow unbounded. This article presents L2GClust, an algorithm to compute local clusterings at each node as an approximation of the global clustering. L2GClust performs local clustering of the sources based on the moving average of each node's data over time: the moving average is approximated using memory-less statistics; clustering is based on the furthest-point algorithm applied to the centroids computed by the node's direct neighbors. Evaluation is performed both on synthetic and real sensor data, using a state-of-the-art sensor network simulator and measuring sensitivity to network size, number of clusters, cluster overlapping, and communication incompleteness. A high level of agreement was found between local and global clusterings, with special emphasis on separability agreement, while an overall robustness to incomplete communications emerged. Communication reduction was also theoretically shown, with communication ratios empirically evaluated for large networks. L2GClust is able to keep a good approximation of the global clustering, using less communication than a centralized alternative, supporting the recommendation to use local algorithms for distributed clustering of streaming data sources.

Abstract
FLUX is a platform for dynamically reconfigurable crowd-sensing using mobile devices like smartphones and tablets, programmed under a notion of region-based sensing. Each region is defined by a set of physical constraints that determine the sensing scope, e.g., based on device position or other environmental variables, plus a set of periodic tasks that perform the actual sensing. The resulting behavior is inherently dynamic: as a device's state changes, e.g., moves in space, it enters and/or leaves different regions, thereby changing the set of active tasks; moreover, regions can be added, deleted, and reprogrammed on-the-fly. FLUX makes use of a domain-specific language for sensing tasks that is compiled into abstract bytecode, later executed by a low-footprint virtual machine within a device, guaranteeing runtime safety by construction. For region/task dissemination, FLUX employs a broker that holds a changeable region configuration plus gateways that mirror the configuration throughout different network access points to which devices connect. Sensing data is streamed by devices to gateways and then back to the broker. Live or archived data streams are in turn fed by the broker to data-processing clients, which interface with the broker using a publish/subscribe API. We conducted two case-study experiments illustrating FLUX: a single-region deployment to monitor WiFi signal quality, and a multi-region deployment to monitor noise, temperature, and places-of-interest based on device movement.

Abstract
We present RAMBLE(1), a framework for georeferenced content-sharing in environments that have limited infrastructural communications, as is the case for rescue operations in the aftermath of natural disasters. RAMBLE makes use of mobile edge-clouds, networks formed by mobile devices in close proximity, and lightweight cloudlets that serve a small geographical area. Using an Android app, users ramble whilst generating geo-referenced content (e.g., text messages, sensor readings, photos, or videos), and disseminate that content opportunistically to nearby devices, cloudlets, or even cloud servers, as allowed by intermittent wireless connections. Each RAMBLE-enabled device can both produce information; consume information for which it expresses interest to neighboors, and; serve as an opportunistic cache for other devices. We describe the architecture of the framework and a case-study application scenario we designed to evaluate its behavior and performance. The results obtained reinforce our view that kits of RAMBLE-enabled mobile devices and modest cloudlets can constitute lightweight and flexible untethered intelligence gathering platforms for first responders in the aftermath of natural disasters, paving the way for the deployment of humanitary assistance and technical staff at large.

Abstract
Edge computing is a hot research topic given the ever-increasing requirements of mobile applications in terms of computation and communication and the emerging Internet-of-Things with billions of devices. While ubiquitous and with considerable computational resources, devices at the edge may not be able to handle processing tasks on their own and thus resort to offloading to cloudlets, when available, or traditional cloud infrastructures. In this paper, we present JAY, a modular and extensible platform for mobile devices, cloudlets, and clouds that can manage computational tasks spawned by devices and make informed decisions about offloading to neighboring devices, cloudlets, or traditional clouds. JAY is parametric on the scheduling strategy and metrics used to make offloading decisions, providing a useful tool to study the impact of distinct offloading strategies. We illustrate the use of JAY with an evaluation of several offloading strategies in distinct cloud configurations using a real-world machine learning application, firing tasks can be dynamically executed on or offloaded to Android devices, cloudlet servers, or Google Cloud servers. The results obtained show that edge-clouds form competent computing platforms on their own and that they can effectively be meshed with cloudlets and traditional clouds when more demanding processing tasks are considered. In particular, edge computing is competitive with infrastructure clouds in scenarios where data is generated at the edge, high bandwidth is required, and a pool of computationally competent devices or an edge-server is available. The results also highlight JAY's ability of exposing the performance compromises in applications when they are deployed over distinct hybrid cloud configurations using distinct offloading strategies.

Abstract
We present a model for measuring the impact of offloading soft real-time jobs over multi-tier cloud infrastructures. The jobs originate in mobile devices and offloading strategies may choose to execute them locally, in neighbouring devices, in cloudlets or in infrastructure cloud servers. Within this specification, we put forward several such offloading strategies characterised by their differential use of the cloud tiers with the goal of optimizing execution time and/or energy consumption. We implement an instance of the model using Jay, a software framework for adaptive computation offloading in hybrid edge clouds. The framework is modular and allows the model and the offloading strategies to be seamlessly implemented while providing the tools to make informed runtime offloading decisions based on system feedback, namely through a built-in system profiler that gathers runtime information such as workload, energy consumption and available bandwidth for every participating device or server. The results show that offloading strategies sensitive to runtime conditions can effectively and dynamically adjust their offloading decisions to produce significant gains in terms of their target optimization functions, namely, execution time, energy consumption and fulfilment of job deadlines.