Software aging is a kind of objective phenomenon with the decrease of the performance and increase of failure rate of the software system. It is a severe test of the reliability of the software system. However, the time point of software aging is uncertain, and a new method based on non stationary time series modeling is proposed for the problem of the current software aging detection, analysis, and the difficulty of the problem of software rejuvenation. In order to provide a more accurate reference for the software rejuvenation, the key parameters of the software running process are established by using the non-stationary time series. An experimental platform, which is based on the current popular Apache server software, is presented. Modeling results show that the average prediction error of the key parameters of software aging is within 4%, which can describe the phenomenon of software aging, and provides a more accurate reference for the actual server.