Research On Load Balancing Of Task Scheduling In Cloud Service System

Along with the progress of science, technology change rapidly, cloud computing has gotten more and more attention, has become a hot topic in the field of computer science. Cloud computing takes virtualization technology, so in this way all kinds of resources will be virtualized into resource pool in the cloud system in order to be managed and offer services to users,thus forming a commercial mode “payment based on the need and amount” which is users-centered. How to efficiently assign the tasks submitted by users and the rational allocation of resources in the pool are the key and difficult points in the cloud computing.It must make sure that users spend less time and money,and the system maintain ints relatively balanced state.Therefore, this thesis targets load balancing as the task scheduling optimization objective. Research work was done as follows:(1) This paper introduces the concept, characteristics, target and task scheduling model of cloud computing task scheduling. Task scheduling model is divided into four categories: centralized scheduling, hierarchical scheduling, distributed scheduling and diffusion scheduling. Based on the diffusion scheduling, the static load model and dynamic load model are introduced in this paper.(2) Under the condition of no task server system input and output, the static load model does balance operation on the initial task,and the entire server system achieves load balancing.This paper focuses on the impact of diffusion scheduling on dynamic load balancing model.(3) In the fully connected server system, when the input and output of the server system is balanced, that is, the input task quantity of the server system is equal to the output task, the server system can achieve the load balance;When input and output is not balanced in the server system,by using the Poisson distribution function to get the task input, system server is unable to achieve the full dynamic load balancing, and experiment to verify that it could achieve better balancing when the average of Poisson distribution function equals to the number of output task.(4) In the imperfect connected server system, the whole system can reach the dynamic load balance when all of the input tasks equal in the server system; In the server system, each input of server is not equal, the server system does not reach full dynamic load balancing, but on the basis of this, using optimization strategy: increase the channel or increase the load transfer coefficient can optimize dynamic load balancing; When input and output is not balanced in the server system,by using the Poisson distribution function to get the task input, system server is unable to achieve the full dynamic load balancing, and experiment to verify that it could achieve better balancing when the average of Poisson distribution function equals to the number of output task.