Analyzing the impact of power subsystem failures and checkpoint mechanisms on availability of cloud applications

The widespread adoption of cloud computing has been driven by technological advantages, such as pay-as-you-go plan, scalability and high availability. From cloud service providers point-of-view, managing all these high customer expectations is a growing and critical challenge. This paper examines the relationship among energy infrastructure architectures, checkpoint mechanisms, and availability of software applications in a cloud data center. We consider four power architectures, three checkpoint mechanisms (cold, warm, and hot) with different sizes of files, and two applications (Video on Demand and digital library). For that, we propose a set of stochastic models (based on Petri Net and Reliability Block Diagram) and perform sensitivity and availability analysis. From results, we note that the power architecture has a great impact on application availability while the checkpoint mechanism only impacts with the application has a low mean time to failure value.