Research Of Software Multi-level Rejuvenation Method Based On Prediction

With the development of network technology, Web application system plays an increasingly vital role and needs to provide the uninterrupted service to users, during long time of operation software aging phenomenon will happen. Software aging will seriously affect the performance of Web application system, interrupt the services to users and cause enormous economic loss, so that the software aging phenomenon has become a problem to be solved in demand. Software rejuvenation, which is a proactive maintenance technology, is an efficient method to tackle software aging problem. System will be rebooted when performance degrades to certain degree, so that the resources can be released and performance can be recovered. There are two important researching points about software rejuvenation. One is the determination of rejuvenating time; another is the determination of rejuvenation objects. Traditional software rejuvenation method usually uses the method based on threshold to determine rejuvenation time, namely periodically monitor system performance, when the performance degrades to certain degree, the rejuvenation operation will be executed. But this method lacks accurate judgment of threshold and will delay rejuvenation time. Moreover, traditional software rejuvenation method only involves two rejuvenation levels, respectively system level and application level. This method does not fit for Web application system whose structure and environment is complex, and will make huge effect on service quality and reliability of Web application system.According to the problems mentioned above, a software multi-level rejuvenation method based on prediction has been proposed. First of all, the structure complexity and environment complexity of Web application system has been analyzed, and rejuvenation levels of Web application system has been divided into four levels, including OS level, middleware level, application level and component level. Moreover, the rejuvenation process and key problems of software multi-level rejuvenation method based on prediction has been introduced in detail, including the classification of rejuvenation level, the collection of system performance metrics, the prediction of system performance degrading trend and determination of component-level rejuvenation strategy. Secondly, a prediction method of software aging trend based on decision tree has been proposed and the whole prediction process has been introduced, including the discretization of system performance metrics data, the reducing of system performance metrics sets and building of prediction model. Thirdly, a component-level rejuvenation strategy based on RBF neutral network has been proposed, which includes the determination of rejuvenation correlation according to coupling relationship of components, the building of rejuvenation group of components through rejuvenation correlations, the calculating of rejuvenation priority of components according to priority metrics using RBF neutral network and the building of components rejuvenation chain.Finally, experiment and case study has been executed in order to verify the effectiveness of proposed method of software multi-level rejuvenation based on prediction. Some aging-related errors has been injected into Web application system to simulate software aging and the real environment, and then the system performance metrics data has been collected through performance monitoring tools. Next, the prediction model of software aging trend has been built using decision tree to predict the degrading trend of system performance. Eventually, components chain has been built using RBF neutral network to rejuvenate components. The experiments results show that the prediction accuracy of proposed prediction model of software aging trend is high, which can guarantee the effectiveness of rejuvenation operation and the proposed component-level rejuvenation strategy is effective and can reduce rejuvenation cost and improve the availability and reliability of Web application system.