Posted to the Linux kernel mailing list this week was a new patch series working on scalability enhancements to the MD RAID5 software RAID code. Up to a 10~17% improvement was observed in some configurations with these RAID5 scalability patches.

Hiroshi Nishida posted the eight patches that are "low risk" changes for enhancing MD RAID5, especially on systems with many CPU cores and many disks as part of the RAID array. One of the primary focuses was on reducing the per-stripe and stripe-cache contention.

Benchmarks provided by Hiroshi Nishida in the patch cover letter are shwoing neutral to some very nice improvements particularly in the RAID5 stripe-handling worker thread performance:

He further elaborated on the findings:

"At the default single handling thread (group_thread_cnt = 0) the series is neutral (no regression). As worker threads are added the gain grows, peaking broadly around group_thread_cnt = 4 at roughly +10-17% across the whole mix; at gtc = 8 the write-heavy workloads keep gaining while the read-heavy high-concurrency case has saturated. (Per-run cv was <1% except the random-write test, ~5-9%, from a cold first run.)"

Those with many-core, many-disk RAID5 array Linux setups can check out the optimization patch series on [1]the LKML where it's awaiting code review.



[1] https://lore.kernel.org/lkml/20260624155452.211646-1-nishidafmly@gmail.com/