New Patches Speed-Up Linux's Accounted Kernel Memory Allocations By ~30%

A set of patches posted last week can improve the Linux kernel's performance of accounted kernel memory allocations by around 30%.

Developer Roman Gushchin posted the set of five patches that can greatly enhance the performance of accounted kernel memory allocations by 30~34% based on a micro-benchmark, but still a ways behind the speed of having kernel memory accounting disabled entirely: the kernel patches remove "the majority of the overhead" for the root memory cgroup while for a user cgroup it's halved.

Gushchin explained of his work:

"The main idea is to get rid of unnecessary memcg to objcg conversions and switch to a scope-based protection of objcgs, which eliminates extra operations with objcg reference counters under a rcu read lock."

See this patch series if interested in the effort to speed-up kernel memory accounting.