7.5.5. cgroups

Table of Contents

Overview and Concepts

Control Groups (cgroups) are a kernel feature that lets you partition and limit a machine’s resources among groups of processes. They are heavily used by containers, systemd, and schedulers.

At a high level, cgroups provide:

Accounting – measure how much CPU, memory, I/O, etc. a group of processes uses
Limiting – cap resources (e.g. max memory, CPU share)
Prioritization – give some workloads more CPU or I/O bandwidth
Isolation – keep one workload from affecting others as much

Unlike namespaces (which provide isolation of views like PID, mount, net), cgroups control resource usage and distribution.

Cgroup v1 vs cgroup v2

There are two major generations:

cgroup v1

Multiple independent hierarchies (one per controller or group of controllers)
Different controllers (cpu, memory, blkio, etc.) can have different trees
Widely used historically, still present on older or compatibility-focused systems

cgroup v2

Single unified hierarchy
Consistent behavior across controllers
Better semantics and delegation model
This is what modern distributions and container stacks favor

Many systems run in a hybrid mode: a cgroup v2 unified hierarchy plus some v1 controllers. For new design and internals understanding, focus on v2.

7.5.5. cgroups

Overview and Concepts

Cgroup v1 vs cgroup v2

Comments

Where to Move