Loading…
Thursday, September 28 • 2:45pm - 3:20pm
如何在大型集群中加速 Pod 的启动? | How Can Pod Start-up Be Accelerated on Nodes in Large Clusters? - Paco Xu, DaoCloud & Byron Wang, Birentech

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
这个想法来自我最近写的博客 Kubernetes 1.27:加速Pod启动的更新。 这是一个集群管理员可能面临的常见问题。 本次会话将向您展示Pod启动的过程以及如何加速Pod的启动。

  1. API:
    控制器管理器创建Pod的时间,
    KCM:PV和PVC绑定以及Webhooks。 
  2. 调度:
    GPU拓扑感知,节点负载感知。
  3. 来自kubelet方面的节点级别:
    镜像拉取,Sidecar,API QPS和Burs,事件驱动PLEG,
    限流,磁盘和卷驱动程序,静态CPU策略,容器运行时。
  4. GPU管理:
    拓扑不仅仅是NUMA,共享和监控。
  5. 数据负载:
    数据预加载和本地存储或分布式存储。
  6. 可观察性:
    如何检查为什么Pod启动缓慢?

The idea came from my recently written blog Kubernetes 1.27: updates on speeding up Pod startup. This is a common issue that cluster administrators may face. This session will show you the process of pod startup and everything about how to speed up the startup of pods.

  1. API:
    the creation time of pods by controller-manager,
    KCM: PV & PVC binding and webhooks. 
  2. Scheduling:
    GPU Topology-aware, node load aware 
  3. Node level from the kubelet side:
    Image Pulling, Sidecar, API QPS & Burs, Event-Driven PLEG, Throttling, Disk and Volume driver, Static CPU Policy, Container Runtime 
  4. GPU Management:
    Topology not only NUMA, Sharing, and Monitoring 
  5. Data load:
    data preload & local storage or distributed storage 
  6. Observability:
    How to check why the pod starts up slowly?

Speakers
avatar for 徐俊杰 Paco

徐俊杰 Paco

Lead of Open Source Team, DaoCloud
Paco is a kubeadm maintainer and an active kubernetes contributor, and he mainly works on SIG-Node & SIG-Cli/SIG-Testing. Paco is currently the leader of the open-source team in DaoCloud, KCD Chengdu 2022 organizer, and a speaker in KCD Shanghai, Kubecon EU 2023, and Kubecon China... Read More →
avatar for Byron Wang

Byron Wang

Staff Engineer, Birentech
Free my soul



Thursday September 28, 2023 2:45pm - 3:20pm CST
2层 会议室 1 | 2F Room 1
  运维+性能 | Operations + Performance