A number of recent studies have reported the phenomenon of “software aging”, characterized by progressive performance degradation and/or an increased occurrence rate of hang/crash failures of a software system due to the exhaustion of operating system resources or the accumulation of errors. To counteract this phenomenon, a proactive technique called 'software rejuvenation' has been proposed. It essentially involves stopping the running software, cleaning its internal state and/or its environment and then restarting it. Software rejuvenation, being preventive in nature, begs the question as to when to schedule it. Periodic rejuvenation, while straightforward to implement, may not yield the best results, because the rate at which software ages is not constant, but it depends on the time-varying system workload. Software rejuvenation should therefore be planned and initiated in the face of the actual system behavior. This requires the measurement, analysis and prediction of system resource usage. In this paper, we study the development of resource usage in a web server while subjecting it to an artificial workload. We first collect data on several system resource usage and activity parameters. Non-parametric statistical methods are then applied for detecting and estimating trends in the data sets. Finally, we fit time series models to the data collected. Unlike the models used previously in the research on software aging, these time series models allow for seasonal patterns, and we show how the exploitation of the seasonal variation can help in adequately predicting the future resource usage. Based on the models employed here, proactive management techniques like software rejuvenation triggered by actual measurements can be built.
