earticle

영어

Mobile cloud computing, which comes up in recent years, is a new computing paradigm. It enables people to access remote clouds by mobile device, even to build mobile micro-cloud(MuCloud) with mobile device to provide lightweight service. Despite extensive studies of task scheduling in wired cloud, effective scheduling in mobile cloud still remains challenges:1) Unreliable wireless connection and dynamic join and quit of MuCloud often result in decreased reliability of scheduling; 2) As the process capacities of wired clouds and MuClouds vary greatly, it is hard to achieve load balancing; 3) During moving, tasks, such as traffic navigation, may be scheduled consecutively by mobile users as space-time changes. Such application scenarios often incur makespan accumulation which impairs user experience, even causes system crash. Our work aims at such problems. We firstly illustrate the reason for selection of makespan and load balancing as two key performance indicators for task scheduling in the proposed architecture of mobile cloud which integrates MuClouds. Then after introduction to Monte Carlo method, degenerated Monte Carlo estimate is defined and a scheduling algorithm based on degenerated Monte Carlo estimate (DMCE) is presented. With extensive simulation experiments, the two above-mentioned indicators of task scheduling using different algorithms including DMCE, Max-Min, Min-Min and IGA are compared and evaluated. Accumulative effect and relative load are introduced to measure scheduling performance. The experimental results show that: 1)Compared with other algorithms, DMCE achieves smallest makespan on average when scheduled respectively; 2) DMCE has least accumulative effect when task sets scheduled consecutively, which makes makespan of a task set hardly relevant to the order of scheduling; 3)Among these algorithms, DMCE outperforms others in keeping relative load balancing by assigning tasks to clouds in proportion to each cloud’s process capacity.