Research On Task Scheduling Algorithm For Mobile Device Cloud

With the development of cloud computing and using of mobile devices, mobile cloudcomputing is developing very fast. But it has an indispensable defect, namely when mobiledevices communicate with cloud, there will be a huge time delay, and mobile devices willconsume a lot of unnecessary electric energy. To solve these problems, researchers came upwith Mobile Device Cloud (MDC).MDC refers to a small cloud composing of a group of mobile devices. These devicescooperate with each other to finish tasks. That means mobile device could transfer its task to anearby device and this device would undertake this task. This pattern will help devices to savemuch energy and also reduce time delay. Researchers prove that transferring task to anotherdevice in the group can save energy and time, and this method can save time and energy up to50%and26%respectively comparing to doing tasks only locally. So proposing MDC is verymeaningful. Our research work focus on how to dispatch tasks in order to extend lifecycle ofmobile devices in the group and improve the throughput at the same time.First, we introduce the development of cloud computing, Cloudlet and MDC, andanalyze the weakness of current researches. Then we analyze related works, including cloudcomputing, Cloudlet, MDC architecture and the difference between them. And we alsointroduce several classical task scheduling algorithms and compare them.Second, we illustrate the overall research plan, and emphasis on task pretreatment,namely research on how to decompose tasks to subtasks. Meanwhile, we consider about thesize of subtasks, in order to reduce the energy cost and enhance the throughput whentransferring these tasks. In the task pretreatment part, we construct task structure graph byanalyzing tasks, and construct independent units based on this. Finally, we construct businesslogic units and get the subtasks.Third, we focus on task scheduling algorithm based on mobile device cloud, includinginitial task scheduling algorithm based on genetic algorithm and follow-up dynamicscheduling algorithm. As for the dynamic scheduling algorithm, we come up with importantmodel, including dynamic device-cloud link model and dynamic task scheduling systemmodel. Then we illustrate dynamic scheduling algorithm based on the previous models.Finally, in the experiment part, we show the result of deco mposing tasks, and prove theperformance of our strategy of decomposing tasks, initial task scheduling algorithm andfollow-up dynamic scheduling algorithm. Our strategy of decomposing tasks and taskscheduling algorithms can effectively reduce energy consumption and enhance throughputunder MDC environment.