Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn how to achieve zero downtime hypervisor kernel updates using kexec while maintaining stateful workloads with modern accelerators in this conference talk from the Linux Plumbers Conference. Discover the challenges of restarting nodes running stateful workloads during infrastructure software upgrades, particularly when applications utilize GPUs, RDMA NICs, and NIC stateful flow accelerators. Explore the innovative approach of bifurcating the mlx5 driver stack's physical function privileges into distinct functions for hypervisor and switchdev offloads. Understand how moving a minimal mlx5 driver to user-space using a generic vfio-pci driver enables delegation of vport management to separate functions. Examine the technical implementation details of this driver bifurcation and switchdev extension that reduces downtime from several minutes to tens of milliseconds or zero downtime for live hypervisor updates while preserving application state in isolated VMs.
