Accelerating task completion in mobile offloading systems through adaptive restart

Mobile application offloading is an efficient technique to unload the burden of intensive computation from thin clients to powerful servers. In a mobile offloading system, cloud computing is utilized to complete some heavy tasks which are migrated from resource-constrained mobile devices to the Cloud. To assure system performance, the quality of the wireless network connection plays an important role. In previous work we experimentally explored the impact of packet loss and delay in wireless networks on the completion time of an offloading task. We investigated a local restart mechanism to mitigate these effects. In the presence of unreliable communication, once the waiting time for the response of a cloud server exceeds a given threshold, exploiting the local resources of a mobile client can accelerate the task completion. In this paper, we upgrade the restart mechanism by allowing several offloading retries before a job eventually is locally restarted and finally completed in the client device itself. This is an adaptive restart scheme which aims first at completing the job using restart with offloading. If several successive offloading attempts fail the job is completed locally. Adaptively selecting the right retry threshold and automatically restarting at the appropriate moment can balance out undesired effects. This paper extends Wang and Wolter (Proceedings of the 6th ACM/SPEC international conference on performance engineering. ACM, pp 3–13, 2015 ) by adding an adaptive retry scheme, a mathematical derivation of the optimal limit for offloading attempts so as to minimize the task completion time using a greedy method, and by the results of a practical evaluation study which shows the efficiency and benefits of the adaptive restart scheme.