vdpa-blk: Unified Hardware and Software Offload for virtio-blk
summary
Proposal builds a unified virtio-blk offload using the vDPA framework, reusing QEMU and kernel code so the same stack supports hardware accelerators and a software device in-kernel; a vdpa-block simulator merged upstream is already available (a ramdisk) and an io_uring passthrough prototype showed promising IOPS between vhost-block and direct io_uring, using polling and io polling. Design adds an automatic fast/slow path switching via a shadow virtqueue to let QEMU regain control for image features and live migration.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:09Goals and benefits of vdpa-blk
- 2:17virtio-blk request path
- 3:31vhost acceleration
- 5:12vhost-bik vsio_uring passthrough
- 6:37virtio Data Path Acceleration (VDPA)
- 8:48VDPA block devices
- 9:44VDPA block software device
- 12:53QEMU auto-switching, fast and slow path
- 14:45Runtime switching between fast and slow path
related talks
24mvDPA Support in Linux Kernel
30mQEMU-Storage-Daemon and libblkio: Exploring New Shores for the QE...
21mVirtio hardware accelerators and other news
31mHardware Friendly Vhost vDPA: Towards an Efficient and Migratable Device Model
21mVirtio Devices Emulation in SPDK Based on Vfio-User Protocol
36mShared Virtual Addressing for High Performance Arm Infrastructure Platforms
34mublk virtual block devices in user space
13mRosenBridge: A Framework for Enabling Express I/O Paths Across the Virtualization...