面向云计算平台的层次化性能问题诊断方法

摘要

提出一种层次式在线性能问题诊断方法——Magnifier,将用户请求执行路径划分为构件层、模块层和方法层3个层次,自顶向下逐层定位问题根源.基于阿里巴巴云计算平台的实验结果表明,Magnifier能够在调用方法种类多、执行路径复杂的条件下,快速定位出性能降级的主要原因.%A hierarchical diagnosis approach, namely Magnifier, was proposed, which models the execution path graph of a user request as component layer, module layer and function layer, and detects anomalies from higher layer to lower layer separately. Extensive experiments were conducted on the Alibaba cloud computing platform. The results indicate that, under the conditions of large volume of data and high complexity of execution paths, Magnifier can accurately and efficiently locate the prime causes of performance degradation.