Research On Reliability Improvement Method Of Cloud Computing System Based On System Level Modeling Theory

With the continuous development of information technology,the demand for computing power in all walks of life is increasing.As a popular computing technology,cloud computing has been widely used.With the increasing number of users,the scale of data center is expanding rapidly,resulting in frequent failures of cloud computing system and huge losses.Therefore,how to ensure the reliability of cloud computing system has become an urgent problem to be solved.Aiming at the reliability of cloud computing system,this thesis proposes some effective fault-tolerant methods to improve the reliability of cloud computing system and reduce or eliminate the impact of errors and faults.1.Aiming at the problem of resource allocation in cloud computing system,this thesis proposes a comprehensive model of cloud resource optimal allocation considering reliability.Auto regressive integrated moving average(ARIMA)model is used to predict the task arrival rate of cloud computing system.Based on queuing theory modeling,an optimal resource allocation model with the goal of minimizing energy consumption under the constraints of service reliability and performance(i.e.response time)is obtained.The experimental analysis shows that in the process of resource allocation of cloud computing system,considering the reliability of response time is very important to ensure the quality of user service.2.Using the fault tree analysis method,this thesis constructs the reliability model of virtual machine system based on replication backup,and puts forward a reliability guarantee mechanism of service level agreement(SLA)based on this model.By looking for the appropriate recovery time point,the system reliability is always maintained above the threshold set by the user.3.Considering that different virtual machines have different fault tolerance requirements(i.e.different virtual machines need different numbers of copies)and compatibility requirements(i.e.some virtual machines and their copies cannot be put into some physical machines due to software or platform incompatibility),a backup virtual machine placement(BVMP)problem under constraints is proposed.BVMP studies how to place virtual machine copies in the cloud data center to minimize the number of physical machines storing virtual machine copies,so as to obtain the least energy consumption.This thesis proves that the BVMP problem is NP hard,and designs an integer linear programming(ILP)algorithm to solve it.ILP method needs a lot of time to calculate,so some efficient and scalable heuristic algorithms are designed to solve the backup virtual machine placement problem.The experimental results show that the ILP based algorithm is superior to the most advanced virtual machine replica placement methods in a wide range of network dynamics.All the proposed backup virtual machine placement algorithms can shut down a large number of physical machines to save energy consumption in the cloud data center.