面向微服务的容器伸缩策略

A CONTAINER SCALING STRATEGY FOR MICROSERVICES

尽管微服务架构可以调整容器的数量,但是仍然存在一个新问题,即在服务的负载压力突然增加或减少时,如何实时准确、快速地调整服务/容器数量。针对该问题,使用时间序列预测模型ARIMA根据最近一个时间区间内pod的负载数据,预测下一个时间区间内的负载。使用预先训练的XGBoost模型,根据预测的负载压力结合pod中cpu和内存使用率预测下一时间窗口内所需的微服务/容器数量。对比kubernetes的HPA算法,该方法能在保证服务质量QoS的同时降低了系统资源的使用。

Although the microservice architecture can adjust the number of containers,there is still a new problem.That is,when the load pressure of the service suddenly increases or decreases,how to adjust the number of services/containers quickly and accurately in real time.To solve this problem,the time series forecasting model ARIMA was used to predict the load in the next time interval based on the load data of the pod in the latest time interval.The pre-trained XGBoost model was used to predict the number of microservices/containers required in the next time window based on the predicted load pressure and the CPU and memory usage in the pod.Compared with the HPA algorithm of kubernetes,the proposed method can reduce system resource usage while ensuring the quality of service.