Prediction-Based Software Availability Enhancement

Abstract

Abstract. We propose a new paradigm for software availability enhancement. We offer a two-step strategy: Failure prediction followed by maintenance actions with the objective of avoiding impending failures or minimizing the effort of their repair. For the first step we present two failure prediction methods: universal basis functions (UBF) and similar events prediction (SEP), which are based on probabilistic analysis. The potential of the presented methods is evaluated by a case-study where failures of a commercial telecommunication platform have been predicted. The second step includes existing maintenance methods fitting the proposed approach and a new recovery strategy called "adaptive recovery blocks". Since system availability enhancement is the overall goal, equations to calculate availability of such a system are given as well.