KVM Dirty Ring - A New Approach to Logging
summary
KVM Dirty Ring replaces the linear dirty-bit bitmap with per-vCPU, mmaped dirty PFN rings that store dirtied page frame numbers; this removes the kernel→userspace bitmap copy, cuts memory footprint, and avoids long mmu_lock re-protection passes. Migration startup for a 128 GB guest reports a ~10× speedup in first iteration. Rings are configurable small buffers, trigger a new VM-exit when full so userspace can recycle entries, and enable finer throttling and better responsiveness; implementation targets Linux x86 and QEMU integration next.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:49Workflow
- 2:10KVM30 Logging
- 4:23KVM Cap Manual Dirty Lock
- 6:10KVM 30 Login Initially Set
- 8:17The Good
- 10:17KVM Dirty Ring
- 12:26KVM Dirty Ring View
- 13:20KVM State Machine
- 15:08Old Dirty Ring vs Dirty Ring
- 15:31Dirty Ring Full Event
- 17:38Side Effect on Auto Converge
- 20:32Conclusions
- 21:33Future works
- 24:46Other solutions
- 25:58References
related talks
33mScaling KVM and Its Community
28mHow We Tackle KubeVirt’s Growth and Scalability
44mUserspace Assisted Scheduling via Sc...
24mNVIDIA Approach for Achieving ASIL B Qualified Linux: minimizing expectations from up...
06mKeynote: Taking Ceph to the Next Level
34mUser Feedback: Shaping the Future of Ceph
09mAspects of Dependable Linux Systems
47mReal-Time Linux: What Is Next?