基于系统性能周期检测的应用服务器软件再生建模与分析

Modeling and analysis of application server rejuvenation based on periodical system performance inspection

软件再生作为一种主动预防性的软件容错技术是解决软件老化问题的主要手段.针对应用服务器系统中存在的软件老化现象,提出一种基于系统性能周期检测的软件再生模型.首先分析应用服务器系统的运行状态和行为特征,给出应用服务器系统单位时间内损失的平均总成本和不可用性的估算公式,并根据最小化单位时间内平均总成本得到两个重要的系统参量:最优的系统检测周期和软件再生周期;然后分析推导单位时间内平均总成本和系统检测周期的边界条件.最后通过仿真实验表明:合理的选择系统检测周期,有计划地对应用服务器系统执行软件再生,可以有效的降低应用服务器系统损失的平均总成本,同时提高系统的可用性和可靠性;并且验证了本文提出的基于周期检测的再生模型,比一般周期再生模型更加提高系统可用性.

Recent studies have reported the phenomenon of software aging. The primary symptoms of this degradation include exhaustion of resources, data corruption and instantaneous error accumulation. Software aging has not only been observed in software used on a mass scale but also in specialized software used in high availability and safety crilical applications. In order to enhance system reliability and prevent systematic degradation or crash, a preventive fault tolerant technique called software rejuvenation had been established in 1995. It involves occasionally stopping the running software, cleaning its internal state and then restarting. At present, software rejuvenation is studied extensively and has many high reliability and availability applications. Considered the stochastic feature of system failure and based on the runtime state of application server, asoftware rejuvenation model based on periodical inspection is set up in this paper. Then the modeling process and performance analysis are given. The theoretic analysis and numeric experiment results show that software rejuvenation model based on periodical inspection can improve systematic availability much higher than the periodical rejuvenation model. This paper can be divided into three main parts. At first, concepts of software rejuvenation and software aging are introduced, and the related works are analyzed. By analyzing the software rejuvenation models which are widely studied nowadays, the shortcomings of those software models are summarized. Second, according to the characteristics of the operational behavior and runtime state of application sever, the periodical inspection method is introduced and the improved software rejuvenation model is set up. Then the optimal systematic inspection interval and software rejuvenation interval are solved from the availability equations, and the bounded condition of expected total cost per unit time and inspection interval is deduced. Finally , a numeric experiment and its result analysis are shown. According to the performance evaluation such as expected total cost and software availability for the application server, there are two parts of the comparative experiment to validate the effectiveness and efficiency of the proposed software rejuvenation model based on periodical inspection.