MSET Performance Optimization for Detection of Software Aging

Software aging [2] is a phenomenon observed in a soft-ware application executing continuously for a long periodof time, where exhaustion of operating system resources(memory leaks), data corruption and numerical error ac-cumulation eventually lead to performance degradation,hang/crash failures or both. To counteract this problem,Huang et al. [2] proposed the technique of software re-juvenation, which involves occasionally stopping the soft-wareapplication,removingtheaccruederrorconditionsandthen restarting the application in a clean environment. Fortime-based rejuvenation policies, which are simpler to im-plement, state restoration is performed at regular determin-istic intervals. For prediction-based rejuvenation, dynamicresource metrics are continuously monitored and rejuvena-tion is attempted only when the onset of aging is deemedhighly probable. The second approach allows the most ef-ficient overall system operation. This approach, however,requires that some performance metrics be identified thatcan be monitored to detect the onset of aging [3].Our objective in this paper is to study the performancetradeoffs in using the Multivariate State Estimation Tech-nique (MSET) for proactive annunciation of software agingin large, Unix-based multiprocessor servers that are used inmission-critical and business-critical e-commerce applica-tions.