摘要
微服务架构可以实现有效的可扩展性、资源隔离和容错隔离,但同时存在级联故障。级联故障由大量微服务间的关联性导致,一旦发生会导致全局性能下降甚至系统崩溃。对此提出一套故障预测方法。在服务网格架构中嵌入级联故障预测组件,对微服务运行的健康程度及微服务间的资源依赖关系进行建模,以获取级联故障参数,并带入GRU神经网络进行故障概率预测。完成了一个面向服务网格的运行时故障预测系统,并进行了仿真实验,验证了该故障预测方法在服务网格中的有效性。
Microservice architecture can achieve efficient scalability, resource isolation and fault isolation. However, cascading failure has become a major obstacle to the microservice systems. Cascading failure is caused by the association between a large number of microservices, and it can cause global performance degradation or even system crashes. A failure prediction method is proposed for cascading failure. A cascading failure prediction component was embedded in the service mesh architecture. The component modelled the health of the microservices and the resource dependencies between the microservices to obtain cascading failure parameters. These parameters were brought into the GRU neural network for failure probability prediction. A runtime failure prediction system for service mesh was completed and the simulation experiment was performed. The experimental results show that the method can effectively predict the cascade failure.
作者
李海飞
徐政钧
Li Haifei;Xu Zhengjun(Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《计算机应用与软件》
北大核心
2021年第11期121-130,共10页
Computer Applications and Software
关键词
微服务
级联故障
服务健康
资源争用
GRU神经网络
Microservice
Cascading failure
Health of service
Resource contention
GRU neural network
