Redundant Eucalyptus Private Clouds: Availability Modeling and Sensitivity Analysis

Cloud computing infrastructures are designed to be accessible anywhere and anytime. This requires various fault tolerance mechanisms for coping with software and hardware failures. Hierarchical modeling approaches are often used to evaluate the availability of such systems, leveraging the representation of complex failure and repair events in distinct parts of the system. This paper presents an availability evaluation for redundant private clouds, represented by RBDs and Markov chains, hierarchically assembled. These private clouds follow the basic architecture of Eucalyptus-based environments, but employing warm-standby redundant hosts for some of its main components. Closed-form equations for the steady-state availability are presented, allowing direct analytical solution for large systems. The availability equations are symbolically differentiated, allowing parametric sensitivity analysis. The results from sensitivity analysis enables system planning for improving the steady- state availability. The sensitivity indices show that failure of the Eucalyptus Cloud Manager subsystem and the respective repair activities deserve priority for maximizing the system availability.