Scheduling for real-time mobile MapReduce systems.

ABSTRACTThe popularity of portable electronics such as smartphones, PDAs and mobile devices and their increasing processing capabilities has enabled the development of several real-time mobile applications that require low-latency, high-throughput response and scalability. Supporting real-time applications in mobile settings is especially challenging due to limited resources, mobile device failures and the significant quality fluctuations of the wireless medium. In this paper we address the problem of supporting distributed real-time applications in a mobile MapReduce framework under the presence of failures. We present Real-Time Mobile MapReduce (MiscoRT), our system aimed at supporting the execution of distributed applications with real-time response requirements. We propose a two level scheduling scheme, designed for the MapReduce programming model, that effectively predicts application execution times and dynamically schedules application tasks. We have performed extensive experiments on a testbed of Nokia N95 8GB smartphones. We demonstrate that our scheduling system is efficient, has low overhead and performs up to 32% faster than its competitors.