負荷履歴を用いたKubernetesにおけるアダプティブスケジューリング

書誌事項

タイトル別名
  • Adaptive Scheduling Based on Historical Resource Usage in Kubernetes

この論文をさがす

抄録

This paper proposes new scheduling methods for a Kubernetes cluster. While Kubernetes is being a very popular container orchestration tool, it does not provide scheduling policies except scheduling pods onto a least or most loaded node. To improve the resource efficiency of a cluster, we propose two scheduling methods, both using historical resource usage data of pods. First, we collect network throughput of a cluster and nodes, and try to place each node onto the node which the scheduled pod has been communicating with the most. Second, we add CPU and memory usage to the first method allowing more general optimization. We use weighted average based on the relative usage of each resource in the cluster and place the pod onto the best node. The method does not work with other schedulers unlike the first method but results in more evenly distributed deployments. Our experiments show the improvements of the stated aspects; 46% and 127% improvement of overall network throughput for the first and the second method respectively, and the second method also shows 43.2% less standard deviation as for CPU usage of each node and 57.5% less as for memory usage.

収録刊行物

詳細情報 詳細情報について

  • CRID
    1390853650463336320
  • NII論文ID
    120007119716
  • NII書誌ID
    AA12746425
  • DOI
    10.15002/00023859
  • HANDLE
    10114/00023859
  • ISSN
    24321192
  • 本文言語コード
    ja
  • データソース種別
    • JaLC
    • IRDB
    • CiNii Articles
  • 抄録ライセンスフラグ
    使用可

問題の指摘

ページトップへ