Practice of Fine-grained Cgroups Resources Scheduling in Kubernetes

Explore a conference talk on fine-grained Cgroups resources scheduling in Kubernetes. Dive into Alibaba's approach to supporting resource scheduling for massive-scale operations, including hundreds of thousands of nodes, millions of containers, and tens of thousands of applications. Learn about their combined scheduling system based on the Kubernetes Scheduler framework and Cgroups controller, which enables dynamic resource limit adjustments, NUMA awareness, and CPU core binding. Discover how this system perceives cgroups level resources and applies dynamic scheduling to specific nodes while allowing certain Pods to bind to specified CPU cores. Gain insights into the implementation of the Cgroup Controller, its practical applications, and the development of a Smart Cgroup Controller. Understand the importance of CPU scheduling, the architecture behind it, and the scheduling framework plugins used. Examine the scheduling algorithm, statistical results, and future developments in this field.

Intro
Agenda
Cgroup Resource Management
What is Cgroup Controller?
Cgroup Controller Implementation
Cgroup Controller in Practice
Smart Cgroup Controller
Cgroup Policy in Practice
Why CPU scheduling is needed ?
Architecture
Scheduling Framework Plugins For CPU Scheduling
Scheduling algorithm
Statistical results
Descheduler
scheduler-plugins
Future works