Deep Dive: SIG VMware
summary
Kubernetes scheduler filters and ranks nodes by predicates and priorities; labels for topology/zones let the scheduler best-effort spread replicas but Kubernetes is unaware of vSphere affinity groups or NUMA unless configured. Use the vSphere cloud provider to expose storage classes, zones, DRS and HA to pods; size VMs to NUMA boundaries (match vCPUs to socket cores, e.g. 8) and avoid odd vCPU counts to reduce remote-memory latency. Enforce CPU as millicores (1/1000 core), set requests/limits, use quotas, and consider vSphere-level resource governance for predictable performance.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 0:17Hui Lou Introduction
- 1:27Agenda
- 3:23Kubernetes Scheduler
- 7:09Modify Scheduler
- 9:11Zones
- 12:25NUMA
- 16:27Virtual Machine
- 18:31Requirements
- 22:10resource management
- 25:27resource quotas
- 27:03nested goals
- 27:31resource management enforcement
- 29:09resource management enforcement hierarchy
- 30:28vSphere DRS
- 31:23Questions
- 31:48High Availability
- 32:14Any Questions
- 33:32vSphere Cloud Provider
- 38:13VMware Technologies
related talks
33mKubernetes VMware User Group Intro: Best Practices for Running on VMware
27mVMware SIG Intro
40mDeep Dive: Kubernetes (VMware) SIG
45mIntro: Kubernetes (VMware) SIG
30mDeep Dive: Scheduling SIG
43mKubernetes in VMware
16mKubernetes VMware UG: What’s New for K8s Users on VMware Infrastructure
35mSIG Scheduling Deep Dive