MSET Performance Optimization for Detection of Software Aging

Software aging [2] is a phenomenon observed in a software application executing continuously for a long period of time, where exhaustion of operating system resources (memory leaks), data corruption and numerical error accumulation eventually lead to performance degradation, hang/crash failures or both. To counteract this problem, Huang et al. [2] proposed the technique of software rejuvenation, which involves occasionally stopping the software application, removing the accrued error conditions and then restarting the application in a clean environment. For time-based rejuvenation policies, which are simpler to implement, state restoration is performed at regular deterministic intervals. For prediction-based rejuvenation, dynamic resource metrics are continuously monitored and rejuvenation is attempted only when the onset of aging is deemed highly probable. The second approach allows the most efficient overall system operation. This approach, however, requires that some performance metrics be identified that can be monitored to detect the onset of aging [3]. Our objective in this paper is to study the performance tradeoffs in using the Multivariate State Estimation Technique (MSET) for proactive annunciation of software aging in large, Unix-based multiprocessor servers that are used in mission-critical and business-critical e-commerce applications.