Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about Google's innovative Live Update mechanism that enables VFIO PCI devices to remain operational during host kernel transitions using Kexec in this 29-minute conference talk from KVM Forum. Discover how this technology addresses the challenge of updating host kernels without requiring live-migration of virtual machines that rely on GPUs or disrupting large-scale Language Model training clusters across numerous hosts. Explore the technical approach to preserving VFIO PCI devices, which allows PCI devices to continue direct memory access and interrupt operations without reset during kernel updates. Understand the significant modifications required to VFIO, IOMMU, and PCI subsystems to achieve this functionality, and gain insights into the development challenges encountered while implementing this solution for maintaining VM operations during kernel live updates.
