Actual Resource Usage-Based Container Scheduler for High Resource Utilization

Abstract

Kubernetes select node and deploy pod based on request to ensure the size of resources for containers with various requirements. In this case, containers are inefficiently managed due to idle resources which are generated by workload configured in various sizes. Therefore, in this study, we propose an Actual Resource Usage-based Scheduler (ARUS), which utilizes the resource usage of each component to perform scheduling to improve the problem of resource waste. ARUS forecasts future resource usage from collected resource usage by utilizing DLinear model. In this case, the optimal node is selected through the scoring for efficient resource utilization (SERU) algorithm. Therefore, ARUS improves resource utilization over conventional kube-scheduler. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.