Performance Estimation And Optimization Of Fault-prone And Retrial Infrastructure-as-a-Service Cloud Computing System

Recently,cloud computing is a hot topic of research institutions and commercial institutions.As computer software and hardware,big data,company business needs developing,cloud computing develops fast.One of hot field of cloud computing is system performance analyzing and optimization.QoS(Quality of Service)is a major factor of customers’ satisfaction,so it must be taken into account in real cloud system.The button level of cloud computing mode is IaaS(Infrastructure-as-a-Service),which directly decides performance of the whole system and expansibility,so it is very important to do research on Iaa S level.We introduce stochastic tools like queuing theory and QBD(quasi birth and death process)to analyze task handling process of IaaS model.Then analyzing system metrics like task delay and task reject rate to illustrate system performance.Those metrics are useful in system optimizing.The process of VM institution may cause fault due to different reasons,so taking fault rate in to account is better.We take task fault and re-submit of tasks into consideration and analyzing the impact of queue capacity,VM multiple ability and some other factors,which cast limitation to the system performance to analyze the cloud computing task handling process as a whole,which illustrates better of real IaaS cloud system.In theory analyzing,we introduce queuing theory,QBD,Markov process to analyze task fault and re-submit.In order to compute arriving rate of tasks and VM institution speed,we record the arrived tasks and VM institution time of a period.Then we use 90% CIs to check the accuracy of theoretical values.The result shows that our theoretical values have high degree of accuracy.Then we analyze the impact on expected response delay and reject rate due to changes of arriving tasks speed、VM creation speed、queue capacity under different numbers of PMs.Based on those values,we do research to decide to best physical machine numbers,queue capacity and some other factors of cloud system to obey SLA.Finally,we introduce simulated annealing algorithm to optimize cloud setting under constrain.In order to show the function,we introduced an example.Based on related research,our work takes task fault and re-submit into consideration,and analyzes the process of cloud computing handling tasks as a whole.In order to analyze task fault and re-submit we analyzed derivation of QBD,and integrate it to our cloud system model.This work offers new modelling and analyzing method for cloud computing performance analysis.