A framework for cluster availability specification and evaluation

Cluster Computing is becoming cost effective and popular for its enormous computational power. High availability features need to be included to ensure that cluster computing environments can provide continuous services. A typical availability modeling method is based on the analytical formalisms such as fault tree, Markov chains, Stochastic Petri Net (SPN), etc. However, people in system design and development may not be familiar with the analytical modeling techniques. This inevitably creates a gap between system designers and reliability engineers. Moreover, the analytical models are still primitive. As a consequence, Markov chain and Petri Net models are often large when the modeled systems are complicated. These large models may be out of the intuitive of modelers, lose the view of the system, and be error prone. We propose a framework that models cluster computing systems' availability based on UML design notations, and evaluates system availability by transforming the UML availability model into corresponding analytical models. The UML-based availability modeling framework is to bridge the gap between the two communities. With our approach, the availability analysis of cluster computing systems can be done at the design stage with ease.