【摘要】 请求级行为分析有助于数据中心应用管理.已有研究中,一类方法要求了解应用内部细节和源代码,因此不适用于数据中心场景,另一类方法基于外部观测应用行为,不能得到精确的分析结果.该文提出一种新的面向数据中心应用的请求级行为分析方法.该方法不要求应用内部细节能够分析请求细粒度性能指标和多项资源消耗.该文贡献主要是:(1)提出了一个结合模型驱动和轻量内核行为跟踪的请求行为分析方法,适用于由黑盒模块构建的数据中心应用;(2)提出了一种利用应用行为预测误差衡量请求分析精度的评价方法.实验表明该方法具有好的分析精度,平均误差小于10%.更多还原

【Abstract】 Request behavior analysis is useful but yet difficult in datacenter scenario.In previous work,one kind of method is application-specific and needs the knowledge about internal details of applications;the other kind lacks analysis accuracy only through external observations on application.In this paper,we propose a novel request analysis method that can accurately estimate request’s fine-grained performance and per-resource consumption without need of application details.Our contributes include: 1) a hybrid method for datacenter application of black boxes that combines model-driven request analysis with a light-weight kernel-level application activities tracing;2) an accuracy evaluation approach that use the error of using request metrics to predict application behavior to quantify the accuracy loss of our analysis method.The experiment demonstrates that our analysis method has acceptable accuracy whose error is lower than 10% at average.更多还原