A Mobile Cloud Computing Middleware for Low Latency Offloading of Big Data

Recent years have witnessed an explosive growth of mobile applications. Thanks to improved network connectivity, it becomes a promising enabling solution to offload computation-intensive applications to the resource abundant public cloud to further augment the capacity of resource-constrained devices. As mobile applications usually have QoS requirements, it is critical to provide low latency services to the mobile users while maintain low leasing cost of cloud resources. However, the resources offered by cloud vendors are usually charged based on a time quanta while the offloading demand for heavy-lifting computation may occur infrequently on mobile devices. This mismatch would demotivate users to resort to public cloud for computation offloading. In this paper, we design a computation offloading middleware which bridges the aforementioned gap between cloud vendors and mobile clients, providing offloading service to multiple users with low cost and delay. The proposed middleware has two key components: Task Scheduler and Instance Manager. The Task Scheduler dispatches the received offloading tasks to execute in the instances reserved by the Instance Manager. Based on the arrival pattern of offloading tasks, the Instance Manager dynamically changes the number of instances to ensure certain service grade of mobile users. Our proposed mechanisms are validated through numerical results. It is shown that a lower average delay can be achieved through proposed scheduling heuristic, and the number of reserved instances well adapts to the offloading demands.