An Improved Kubernetes Scheduling Algorithm for Deep Learning Platform

Most existing deep learning platforms only focus on helping users to start task training quickly, but they tend to ignore the application scenario of multi-team collaboration using one resource pool. In this paper, we propose an improved scheduling algorithm oriented to a multi-tenant model, in which team users are modeled as virtual clusters and cluster load will be monitored regularly. We apply the optimized Kubernetes scheduling algorithm to the Docker-based deep learning platform, our method can ensure the load balance and meet the needs of users.