Research On The Task Scheduling For The Cloud Computing Platform

Recently with the rapid development and widespread applications of information technique and Internet,the demand for computing,storage and various software services is increasing by individuals and enterprises.The cloud computing is a new business mode of on-demand service that enables quick and easy access to all kinds of resource via the Internet.For the cloud computing,the task scheduling is one of the key techniques.How to allocate the tasks subscribed by users to the data center of cloud computing in a reasonable and efficient manner is not only related to the user experience and service quality,but also impacting on the capability of cloud service provider,load balance among server clusters,and the control of operation cost.In this thesis,the independent task scheduling algorithms and the associated task scheduling algorithms for the cloud computing platform are studied and improved.The main work of this thesis is as follows:Firstly,the technical background and knowledge about cloud computing are introduced,and the key techniques of cloud computing platform are analyzed.The current task scheduling algorithms for cloud computing are studied in detail by comparing the advantages and disadvantages of various algorithms in terms of research purpose and methods.Then,for the independent task scheduling,from the perspective of improving service quality,a time aware independent task backfilling scheduling algorithm is proposed based on the backfill algorithm.This algorithm overcomes the problem of virtual machion starvation caused by the single index in the traditional backfill scheduling algorithm.Combined with the estimated completion time and the requested processor cores,the tasks are backfilled and a time-aware load balancing strategy is adopted.Finally,the makespan of the task is increased,and the waiting delay of task queue is reduced while keeping the load balance.Finally,the table scheduling algorithm with lower complexity and good performance is studied for the associated task scheduling.The classical HEFT and CPOP algorithms are improved during the stage of computing tasks weight,which has the problem of large computing and communication cost caused by the insufficient consideration of heterogeneous environment.The standard deviation is used as the coefficient to reflect the heterogeneity calculation difference,and the ratio of outdegree to indegree of each task in the associated tasks is used as the heterogeneous factor of the communication overhead to increase the priority of those tasks with the large ratio of outdegree to indegree tasks.It reduces the communication overhead of subsequent tasks and improves the scheduling length of the associated tasked and the parallel execution.During the phase of virtual machine selection,the task assignment is optimized by selecting the best virtual machine depending on the computational and communication overheads of critical tasks on the critical paths and their successor nodes.Totally it improves the schedule length ratio and parallel execution degree of associated tasks,and works well in the scenarios of large heterogeneous parameters.