Software Aging Prediction and Availability Analysis

Abstract

The 'software aging', is characterized by progressive performance degradation due to the exhaustion of operating system resources. Software rejuvenation should be planned and initiated in the face of the actual system behavior which requires measurement, analysis and prediction of system resource usage. In this paper, we present availability analysis and prediction for software aging in virtualized environment. The difficulty is that it can be due to two or more resources simultaneously involved in the service failure. So, we decide to evaluate the use of powerful Machine Learning algorithms to predict the time to crash (TTC) of a system which suffers from software aging phenomena. We also present time dependent software rejuvenation (TDSR) policy that can be applied with predictable data in virtualized environment. The behavior of the system is represented through a Stochastic Petri Net (SPN) model. Numerical analysis of the system availability is carried out the SHARPE tool simulation.