KVM Dirty Page Tracking

on YouTube ↗25m

" summary

Explains that KVM uses synchronous and asynchronous dirty tracking for snapshots and migration and reports concrete measurements: migrating a 3 TB guest with a 100 MB bitmap shows bitmap copying costing 200 ms and using atomic compare-exchange to copy bitmaps is about 3x slower than memcpy (up to 8x in L1), so performance pain comes from many bitmap layers and atomic memory bus locking. Proposes reducing bitmap layers, avoid atomics for bulk copy by taking a read‑write lock, and using dirty ring, clear log and zero‑copy IO to scale huge VM migration.

AI-written summary · about

" chapters (11)

1. 0:00Introduction
2. 0:17What is Dirty Page Tracking
3. 1:09Synchronous Dirty Tracking
4. 3:36Asynchronous Dirty Tracking
5. 7:48Huge VM Migrations
6. 10:20Summary
7. 12:24Downtime
8. 19:04Atomic operations
9. 20:28Cache mis
10. 21:25KVM solution
11. 23:02QML solution

" related talks

• 25mExploring Intel Sub-Page Protection in Context of Live MigrationManish Mishra, Shivam Kumar · KVM Forum 2022
• 25mXen and XenServer Storage PerformanceFelipe Franciosi · Xen Project User Summit 2013
• 40mAchieving the Ultimate Performance with KVMBoyan Krosnov · SREcon20 Americas
• 19mSpeeding Up VM’s I/O Sharing Host's io_uring Queues With GuestsStefano Garzarella · KVM Forum 2020
• 44mImproving kexec boot timeUsama Arif · Linux Plumbers Conference 2023
• 20mSpeed Up Creation of a VM With Pass Through GPULiang Li · KVM Forum 2020
• 22mQEMU Snaphosts Are So Slow. Really?Denis V. Lunev · KVM Forum 2020
• 15mThe Practice Method to Speed Up 10x Boot-up Time for Guest in Alibaba CloudWeinan Li · KVM Forum 2020