Performance analysis of software rejuvenation

Abstract

Cluster-based systems, a combination of interconnected individual computers, have become a popular solution to build the scalable and highly available Web servers. In order to reduce system outages due to aging phenomenon, software rejuvenation, a proactive fault-tolerance strategy has been introduced into cluster systems. Compared with clusters of a flat architecture, in which all the nodes share the same functions, we model and analyze the dispatcher-worker based cluster systems, which employ prediction-based rejuvenation both on the dispatcher and the worker pool. To evaluate the effects of rejuvenation, stochastic reward net models are constructed and solved by SPNP (stochastic Petri net package). Numerical results show that prediction-based software rejuvenation can significantly increase system availability and reduce the expected job loss probability.