Availability Analysis Of A Server Virtualized System Based On Stochastic Reward Nets

In recent years,the IT enterprises adopt the server virtualization technology as the most common method in IaaS to provide cloud computing services.The physical servers bear the computing and storage services of companies.The security and reliability of physical servers become very important.If the service disrupts due to attacks or its software and hardware failures,it will lead to unpredictable economic losses.Therefore,how to ensure the high availability of the server virtualized system becomes the common topics which the cloud service providers and users care about.The factors which affect the availability of the server virtualized system can be divided into two aspects including hardware and software.The hardware factors contain the hardware and the storage network failures.The software factors contain the software bugs in VM and VMM.These two factors will lead to the decrease of the server virtualized system availability.In response to these unsafe factors,there are some common high availability techniques in a virtualized system,such as software rejuvenation mechanisms both in VM and VMM,VM failover technique,live VM migration technique.This article investigates the effect of combination of these high availability techniques on VM availability in a virtualized system where various software and hardware failures may occur.For each combination,we construct analytic models by using Stochastic Reward Nets.This article analyzes performance indicators such as the steady-state availability and downtime.Sensitivity analysis method is used to analyze the impact of the variables in the models on the availability of the virtualized system.The numerical results of these models indicate that the server virtualized system can achieve the maximum availability by combining the failover and live VM migration technique together and using time-based rejuvenation mechanism for VMM and prediction-based rejuvenation mechanism for VM.These analytic results provide guidelines for deploying and parameter setting of FT and HA techniques in a server virtualized system.