Availability measurement and modeling for an application server

Abstract

Application server is a standard middleware platform for deploying Web-based business applications which typically require the underlying platform to deliver high system availability and to minimize loss of transactions. This paper presents a measurement-based availability modeling and analysis for a fault tolerant application server system - Sun Java System Application Server, Enterprise Edition 7. The study applies hierarchical Markov reward modeling techniques on the target software system. The model parameters are conservatively estimated from lab or field measurements. The uncertainty analysis method is used on the model to obtain average system availability and confidence intervals by randomly sampling from possible ranges of parameters that cannot be accurately measured in limited time frames or may vary widely in customer sites. As demonstrated in this paper, the combined use of lab measurement, analytical modeling, and uncertainty analysis is a useful evaluation approach which can provide a conservative availability assessment at stated confidence levels for a new software product.