A Container Scheduling Strategy Based on Machine Learning in Microservice Architecture

Abstract

Good progress has been made in the microservice technology in recent years due to the development of docker container technology. Although microservice architecture can adjust the number of containers, there is a new problem. That is according to the current load pressure of services, how to adjust the number of containers accurately and quickly in real time, especially when the load pressure of services suddenly increases or decreases. In order to deal with the problem, a container scheduling strategy based on machine learning is proposed in this paper. The paper uses the data set obtained by our experiments to train random forest regression model in advance to predict the required containers of services in the next time window, according to the current load pressure of services. Large amounts of data has been collected to make the data set for the experiment. Based on the data set, random forest regression model has been trained in advance to predict the number of required containers of services in next time window. When adjusting the number of containers to balance load pressure of services, the proposed algorithm saves 50% of the time compared with traditional algorithms. In addition, the accuracy is 10%~38% higher compared with other machine learning algorithms.