Automated Adaptive Restart for Accelerating Task Completion in Cloud Offloading Systems

Abstract

Offloading is a technique that utilises the powerful computation resource of Clouds by migrating heavy computations from thin clients like mobile devices to a remote server. Although task completion in the cloud is usually fast, an unreliable network connection often causes delays or interruptions which level the advantages of powerful resources off. Restart is an efficient method that can under certain conditions reduce the task completion in computer and network systems. In this paper we introduce an automated restart scheme. It aims first at completing the job using restart with offloading. Once the number of offloading attempts exceeds a threshold, the job is completed locally. A key challenge is to identify the optimal limit for offloading attempts as to minimise the task completion time. To address this problem we mathematically derive the expected task completion time under different thresholds and compare results of our analysis.