MobSched: Customizable scheduler for mobile cloud computing

In this paper, we explore how cloud computing techniques can be used on mobile devices. We analyze the reason why Hadoop´s performance is poor in MANET, most notably, relying too much on distributed filesystem, and not aware of mobility and multi-hop nature of MANET. Two ways are proposed to deploy mobile cloud computing in an efficient manner: MobSched, a customizable job scheduler; and a mobile friendly MapReduce framework. These two methods enable developers to use MapReduce programming model in the context of MANET. Theoretical analysis suggests that the proposed framework can improve the performance of MapReduce jobs running on top of MANET, and reduce the energy consumption. Simulation results show that the proposed scheduler, MobSched, which is based on a linear programming formulation, can efficiently optimize multiple objectives such as power and (or) throughput, while being constrained with requirements such as minimum quality of service, and (or) maximum bandwidth usage that has to be met by the system. Comparison with other schedulers such as uniform load balancing, FIFO, and clustering types show that the proposed scheduler performs best when it comes to optimizing for a specific criteria such as total power consumption within reasonable latency.