基于机器学习的微服务系统性能自动检测方法

Automatic Performance Detection Method of Micro Service System Based on Machine Learning

微服务系统性能过低,导致应用软件扩展性较差、可靠性降低以及维护成本增加等问题。为此,提出基于机器学习的微服务系统性能自动检测方法。采用神经网络算法构建执行单元概率矩阵,确定指数级的训练时间,利用分层分析微服务种类,以闵氏距离法明确各执行单元间距离,凭借训练数据集建立自动编码器,构建起始训练数据。通过MySQL数据库、分布式追踪等构建系统,将收集数据存储在数据库内,生成特征规则,最后将异常检测模块与提取的特征规则对比,完成性能检测。实验结果表明方法能够准确检测到系统最大吞吐量、且检测的耗时较短。

The low performance of microservice system leads to the problems of poor scalability,low reliability and high maintenance cost of application software.Therefore,an automatic performance detection method of micro service system based on machine learning is proposed.The neural network algorithm is used to construct the probability matrix of the execution unit to determine the exponential training time.The hierarchical analysis of micro service types is used to determine the distance between the execution units by Min's distance method.The automatic encoder is established by virtue of the training data set to construct the initial training data.Through MySQL database and distributed tracking system,the collected data is stored in the database,and the feature rules are generated.Finally,the anomaly detection module is compared with the extracted feature rules to complete the performance detection.Experimental results show that the method can accurately detect the maximum throughput of the system,and the detection time is short.