Analyzing the impact of power infrastructure failures on cloud application availability

Abstract

The widespread adoption of cloud computing has been driven by perceived advantages in agility and scalability inherent in cloud computing. For hyperscale cloud service providers, managing the high expectations of customers and the damage resulting from service outages is an increasing and critical challenge. Data power failure is a significant risk to cloud service availability. This paper examines the relationship between power infrastructure architectures and software application availability in a hyperscale data center. Using sensitivity analysis, we define four power infrastructure architectures and analyse the impact of power infrastructure failures on applications hosted in a hyperscale data center. We use Stochastic Petri Net (SPN) models to represent the power infrastructure behavior, and Reliability Block Diagrams (RBD) to represent application availability. Results from stationary analysis suggest that the availability gain on a highly redundant architecture varies depending on the application type.