Tecent engineers have been working on addressing long-standing inefficiencies within the Linux kernel scheduler code around over-subscribed virtualized environments.

A set of ten patches were posted today to deal with KVM and scheduler limitations within the Linux kernel around over-subscribed VMs running on the Kernel-based Virtual Machine (KVM). The patches implement a scheduler vCPU debooster and KVM IPI-aware directed yield support. These new mechanisms are designed for minimal overhead and from Tencent’s view is superior to other para-virtualization approaches.

Long story short for end-users/administrators just wondering about the net gain for these new patches on over-subscribed VMs on Linux servers:

And the analysis from the patch cover letter:

“- Gains are most pronounced at moderate overcommit (2-3 VMs). At this level, contention is significant enough to benefit from better yield behavior, but context switch overhead remains manageable.

• Dedup shows the strongest improvement (+47.1% at 2 VMs) due to its IPI-heavy synchronization patterns. The IPI-aware directed yield precisely targets the bottleneck.

• At 4 VMs (heavier overcommit), gains diminish as general CPU contention dominates. However, performance never regresses, indicating the mechanisms gracefully degrade.“

Those interested in this work can find the patch series out for review on the Linux kernel mailing list.