Real Time Processing in Mobile Clouds

With rapid advances in mobile device and cloud computing technologies, a new computing paradigm in which large amounts of data are stored and processed on mobile devices is emerging. Despite the powerful hardware available, mobile devices have limited capacities as they are powered by battery and connected by unstable, low bandwidth, wireless networks. Apache Storm is a scalable platform that provides distributed real-time stream processing paradigm and fault tolerant capability. This paper studies the existing problems of applying Storm to mobile environment, and then proposes a new framework to address these problems with the goal that it would outperform Storm in performance in mobile environment. More specifically, we hope that our framework would reduce processing latency, energy consumption and provide guarantee that processing latency is under certain predefined threshold. Concretely, we formulate the resource allocation and task scheduling optimization problem and propose a heuristic solution to approximate the optimal solution. In our heuristic solution, we generate task scheduling and resource (worker node) allocation strategies according to collected inter-task traffics and latency information of running topologies. Extensive evaluations are performed through proof-of-concept real hardware implementation. Results show that our proposed framework effectively reduces processing latency by up to 50% compared with Storm, also it controls processing latency under certain predefined threshold in abnormal situation.