This is The Way: A Crash Course on the Intricacies of Managing CPUs...
summary
Kubernetes adds CPU Manager (1.8) to give static CPU manager exclusive cores for containers in Guaranteed QoS, with the beta full-pcpu-only option rejecting pods that would share a physical core; Topology Manager (1.16) aligns CPUs and devices but scheduler topology-unawareness causes admission failures and runaway pod creation; Memory Manager appears 1.21/1.22 to close gaps. Interim node solutions include crirm, CPU Pooler and deprecated CMK. Remaining limits: container-level pinning only, eight NUMA zones max, no CPU-type requests, and in-place resizing targets 1.25.
AI-written summary · about
chapters (11)
- 0:00Intro
- 1:08Resource Management in Kubelet
- 6:08CPU Manager Policies
- 7:02CPU Manager Policy Options (Introduced in v1.22, Beta in v1.23)
- 8:54Exclusive CPU Assignment to a pod
- 9:31NUMA Zones: Not for the weak of heart
- 10:06Along Comes Topology Management...
- 11:51Going with the Topology Flow
- 16:16Out of Tree solutions for CPU Management • All rely on turning off all resource management by
- 18:38Community Discussion
- 21:08Working Groups
related talks
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
18mLet There Be Topology-Awareness in Kube-Scheduler!
36mTaming the Beast: Advanced Resource Management With Kubernetes
30mDeep Dive: CRI- RM Based CPU and NUMA Affinity to Achieve AI Task Accele...
24mPropagating Programming Paradigms: Lifting High Performance Compute Into K8s
21mDo Not Disturb Mode: Fine-Tuning Resources for Latency Sensitive Workloads
38mA Tale of Tail Latency: Understanding k18n CPU Requests and Limits for Sust. and Profit
47mCPU & Memory, Requests & Limits, QoS & CFS: The JVM in the Kubernetes Pod