Deep Optimization of VMM Live Upgrade
summary
Replaces the naive reuse of live-migration by keeping vhost-user backends running and inheriting their FDs, adding an upgrade-agreement mode that avoids restarting PCI devices, skipping expensive set/get communications and pretransferring only compact config deltas. For a 64 vCPU, 128 GB VM with 2 multiqueue vhost-user-net and 10 multiqueue vhost-user-blk devices downtime drops from ~900 ms to ~150 ms and packet loss goes to zero; experiments show ~80–90% downtime reduction on smaller and larger VMs.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:37Background
- 1:48Overview of VMM Live Upgrade
- 3:39Downtime Breakdown
- 4:43Optimizations - Insight
- 5:35Optimizations - Transparent to Backends
- 9:07Optimizations - Presave Config
- 11:16Achievements - Downtime
- 12:45Achievements - Packet Loss
related talks
24mSupport for Fast and Reliable VMM Live Upgrades in Libvirt
15mKVM Live Upgrade with Properly Handling of Passthrough Devices
50mOps issues
22mDevice Keepalive State for Local Live Migration and VMM Fast Restart
36mNFVI Upgrades and NFV Migrations with Live Critical Telco
10mTowards a More Efficiently Synchronization in KVM
35mMigrating a live site across the country without downtime
15mKeeping a Hypervisor Fleet Up to Date with Minimal Customer Disruption