Cgroup Slab Memory Controller and Time Namespace
summary
New kernel features back-ported to RHEL8 include a cgroup v2 slab memory controller (merged in Linux 5.9) that replaces per-cgroup per-CPU slab caches with shared slab caches plus an object_cgroup layer to account bytes per object; measured on POWER systems kernel slab for a pod drops from ~3 GB to ~400 MB (≈7×) and per-container slab from 112 MB to 90 MB, with ~17% runtime gain. The time namespace virtualizes CLOCK_MONOTONIC and CLOCK_BOOTTIME via vdso viewer-page offsets for consistent checkpoint/restore, currently implemented on x86_64 and arm64.
AI-written summary · about
chapters (6)
related talks
30mdrm and cgroup memory management - two great tastes
01h 36mLinux Memory Management and Containers
43mExpanding the Memory Limits with CXL Software Stack
22mRestartable Sequences: Scaling Per-Core Shared Memory Use in Containers
25mThe slab allocators of past, present, and future
32mA mechanism to isolate CPU topology information in the Linux kernel -- CPU Namespace
38mUnderstanding and Working with the Cgroups Interface
20mGet More Computing Power by Helping the OS Scheduler