Performance Analysis And Evaluation Of Cloud Computing System

In recent years,with the development of computer technology,cloud computing becomes more and more popular.Through using distributed processing,cloud computing can effectively improve its computing ability.What's more,because of the high augmentability of cloud computing,users can easily extend resource using or release redundant resources basing on their need,and this is conducive to improving resource utilization.With the development of cloud computing,the number of jobs or applications based on cloud computing is increasing in a quite rapid speed,which result in higher and higher requirement for the performance of cloud computing systems.Business requirement drive people to pay more and more attention to the performance of cloud computing systems.In this paper,the performance of cloud computing systems are mainly analyzed in two aspects,one is computing performance,and the other is the network performance.(1)In the aspect of computing performance,a model about the relationships among the jobs,tasks and shared physical resources in the cloud computing system is built.Virtualization technology is the basis of cloud computing,and it makes a large number of virtual machines to run concurrently and to share the resources of the resource pool.However,just because of the resource sharing,performances of some jobs deployed in the cloud computing change along with the variation of other jobs or resources,and this is called as "performance jitter".What's more,jobs deployed in the cloud computing(especially the computing intensive tasks and storage intensive tasks)are usually divided into several tasks,which are allocated to different virtual machines to be executed in a parallel mode.On the basis of these theories,in this paper,a model is built to describe the division of the jobs,the allocation of the tasks and the resource sharing.In this model,the dependency relationships among all the jobs,tasks,shared physical resources in the entire cloud computing system are included,and based on the model,the computing performance of the entire cloud system can be analyzed on the overall perspective.(2)In the aspect of computing performance,based on the built model,this paper further studies the "computing performance jitter" of the cloud system.The division degree of the jobs(which refers to the number of tasks that a job is divided into)determines how close the sharing relationship between the jobs are,then determines the "computing performance jitter"degree of the jobs.The researches are conducted in two scenarios,one is that a new job is inserted into the cloud computing,the other is that a shared physical resource is down.The simulation results show that the"performance jitter" of one job is not only related to the variation of other jobs or resources,but also closely related to average division degree of the jobs.(3)In the aspect of network performance,a MRF(Markov Random Field)model is built to describe the coupling relationship among the data flows in the cloud computing data centers.Since the jobs deployed in the cloud computing system are usually divided into a number of parallel tasks,and there may exist data exchange during the execution process,the execution efficiencies of the jobs are not only related to the computing performance of the cloud computing system,but also related to the network performance of the cloud computing system.A large amount of data flows in the cloud computing data centers share the bandwidth resources of the physical links,and thus result in the complex coupling relationships among them.On the basis of the theory above,in this paper,a model describing the coupling relationship among all the data flows in the entire cloud computing system is built through introducing the MRF(Markov Random Field).And based on the model,the network performance of the entire cloud computing system can be analyzed on the overall perspective.(4)In the aspect of network performance,based on the built model,this paper further studies the stability of the network performance in the cloud computing system.The built MRF model is solved by introducing the BP(Belief Propagation)algorithm.Under the premise that the bandwidth requirement of each data flow and the bandwidth capacity limit of each physical link are fixed,the BP algorithm adjust the bandwidth allocation for all the data flows in each iteration.And the simulation results show that the stability of the network performance in the cloud computing data centers can be continuously improved through the BP algorithm iteration process.