Live-Update: An Update on the Journey to Replace the Hypervisor with Limited Downtime
summary
AWS ships a seven-step live-update that loads a new Xen binary into free RAM, preserves domain page metadata, and resumes guests after transferring a compact live-update with memory blob. The system writes a stream page list of MFNs, schedules pre- and post-save tasklets on CPU0, and jumps to Xen2 with a command‑line pointer. On c5.metal QEMU tests it achieves <1s downtime on c5.metal for an 8GB dom0 and 64GB host; sorting reduced page-list entries from ~1,270 to 9 and brought 730ms to ~670ms.
AI-written summary · about
chapters (7)
- 0:00Introduction to Live Update
- 1:04Why Live Update is needed
- 3:56Live Update process overview
- 15:14Live Update downtime metrics
- 16:17Optimizing memory transfer
- 20:57Alternative: Never pausing domains
- 23:03Conclusion and future work
related talks
20mKVM-devirt: Extending KVM to a Zero-Overhead Partition Hypervisor
27mXen: This Is Not Your Dad's Hypervisor!
27mEvaluate Implementation Options of KVM-based Type1 (or 1.5) Hypervisor
24mYank: Enabling Green Data Centers to Pull the Plug
42mEnhancing Hypervisor and Cloud Solutions Using Embedded Linux
32mHVX: Virtualizing the Cloud
29mChanging the Engine Mid-Flight: Zero-Downtime Ceph Upgrades
32mNotes on Taking KVM-on-KVM Nested Virtualization for a Spin
