基于队列模型的软件老化检测

Software aging detection method based on queuing model

针对传统的用于软件老化检测的方法忽略外部负载对老化的影响而易产生老化误报的问题,同时考虑性能参数与外部负载,提出了基于队列模型的融合外部负载的软件老化检测方法。队列模型输出每种事务在应用服务器中的服务时间,这种度量称为事务的性能"签名"(简记为TPS),以此作为软件老化度量指标,通过TPS的变化检测软件老化。基于TPC-W事务处理系统,设计与实现了包含队列模型的融合外部负载的软件老化检测系统。利用基于队列模型的检测方法在TPC-W测试床上进行软件老化检测得出了如下结论:基于TPS的老化检测可以融合外界负载因素,有效地检测软件老化;并且通过合理选择监测窗口,优化检测效果。基于TPS的检测方法对不同的变化负载类型和性能数据同样可以有效检测软件老化。通过与已有的仅依赖于系统性能数据的软件老化检测方法AR(自回归)比较,基于TPS的软件老化检测误报次数明显低于AR模型。综上所述TPS是一种能够有效地检测软件老化并显著减少软件老化错误报告的鲁棒性的软件老化检测方法。

Regarding the false alarms caused by traditional methods of software aging detection which ignore the effects of ex- ternal load on software aging, both performance parameters and external load are taken into account in proposing a software ag- ing detection method based on queuing model. The service time of each transaction in application server as the output of the queuing model is called the transaction performance signature, abbreviated as TPS. TPS is used as the metric to detect software aging through the changing of them. In this paper, a software aging detection system is designed and developed based on TPC-W transaction processing system, which contains a queuing model. We have got the following conclusions through analyz- ing aging detection with the queuing model in real TPC-W benchmark. Integrating external load, the detection method based on TPS can obtain more effective result. By properly selecting the size of the detection window, higher detecting quality can be at- tained. For different types of external load and different performance parameters, the detection method based on TPS can also get expected results. Compared with Auto-Regressive （AR） model which only depends on original performance parameters, the number of false alarms is significantly lowered using the detection method based on TPS. In conclusion, the robust software ag- ing detection method based on TPS can get effective results and significantly reduce the nseudo aging alarms.