量化评估用户感知的可用性

Quantized Evaluation for User Aware Availability

为了量化分析Web服务失效和性能降级的原因,提出了基于Web集群服务器的用户感知的可用性模型.首先使用Markov生灭过程估算系统状态概率,再利用排队论估算系统稳定状态下的请求丢失概率,最后结合二者建立起用户感知的可用性数学表达式,并以此来量化评估用户感知的可用性对各系统性能参数的敏感性.数值计算结果表明,影响用户感知的可用性的主要因素有2种,即低负载下的平均故障检测时间与平均无故障时间之比值,高负载下的平均故障修复时间与平均无故障时间之比值,同时用户感知的可用性可以通过增加后端节点的个数来改善,当后端节点数增加到一定程度时,该性能将趋于平稳.

To quantitatively analyze the reason of Web service failures and performance degradation, a model of the user perceived availability in cluster based Web servers is proposed. Firstly, the system state probability is estimated by Markov birth-death process, and then the request loss probability in stable system states is estimated by queuing theory, lastly, combining both of them, therefore the expression for the user perceived availability is established to quantitatively evaluate the sensitivity of user perceived availability to different system parameters. The numeri- cal results show that the dominant factors that impact the user perceived availability are the ratio between the mean detecting time and the mean fault-free time in low workload and the ratio between the mean repair time and the mean fault-free time in high workload. Meanwhile, the user perceived availability can be improved by increasing the number of back end nodes, however, when the number of back-end nodes is increased to a certain extent, it becomes steady.