Features/VirtioSCSI

From QEMU
Revision as of 15:39, 11 October 2016 by Paolo Bonzini (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Overview

A virtio storage interface for efficient I/O that overcomes virtio-blk limitations and supports advanced SCSI hardware.

Features

  • /dev/sda inside guest
  • Multiple LUNs per PCI adapter, solves virtio-blk 1:1 scalability problem
  • SCSI passthrough (tape, CD burning, ...)

Status

  • Virtio device specification: part of VIRTIO 1.0
  • Linux guest driver: working
    • Multi-queue / fine-grained per-queue locking
  • QEMU hardware emulation using SCSI emulation: working
    • Added support for WRITE SAME commands
    • Added support for hotplug events
  • QEMU hardware emulation using vhost-scsi: working
  • Libvirt support: working
  • SeaBIOS support: working
  • Tested use include:
    • CD-ROM/DVD burning passthrough
    • Tape passthrough
    • 100s of LUNs (scalability)
    • LUN hotplug (known to work with manual hotplug)

Future tasks

  • Pick up vhost-scsi work again:
    • Port QEMU hw/virtio-scsi.c vhost-scsi support onto latest code
    • add QEMU Object Model (QOM) support to the vhost-scsi device
    • Port LIO vhost-scsi code onto latest lio.git code
    • Update vhost-scsi to implement latest virtio-scsi device specification
    • Ensure vhost-scsi I/O still works
    • Design libvirt integration for LIO
  • QEMU SCSI target:
    • Add support for rerror/werror to scsi-generic
  • Complete virtio-scsi spec implementation:
    • Add support for asynchronous media change notifications
  • Use cases to flesh out:
    • Multipath inside guest
  • Performance analysis
    • Compare against virtio-blk I/O (rough numbers: 6% slower on iozone with a tmpfs-backed disk)
    • Compare passthrough performance against bare metal host SCSI
    • Measure scalability and where the bottlenecks are (full virtqueues? locks?)

Specifications

Links