Jump to: navigation, search

Virtio-scsi-for-bdm

Revision as of 15:27, 1 April 2014 by Trump.Zhang (talk | contribs) (Created page with "== Add virtio-scsi bus support for block device mapping == The corresponding BP is https://blueprints.launchpad.net/nova/+spec/add-virtio-scsi-bus-for-bdm ===Virtio SCSI===...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Add virtio-scsi bus support for block device mapping

The corresponding BP is https://blueprints.launchpad.net/nova/+spec/add-virtio-scsi-bus-for-bdm

Virtio SCSI

VirtIO SCSI is a new para-virtualized SCSI controller device for KVM instances. It has been designed to replace virtio-blk, increase it's performance and improve scalability. The interface is capable of handling multiple block devices per virtual SCSI adapter, keeps the standard scsi device naming in the guests (e.x /dev/sda) and support SCSI devices passthrough.

Goals

  1. Boot from a cinder volume with virtio-scsi bus type
  2. Attach a volume to instance using virtio-scsi bus type

Use cases

  1. Improve performance in I/O-intensive applications
  2. Support virtio-scsi passthrough, and with "multi-attached-volumes" ([1]) together, can be used to support MSCS or Oracle RAC to run in cloud servers

Libvirt Driver Support

When using a volume with "scsi" bus type and not specify a scsi controller explicitly in the domain xml, libvirt will automatically generate a "lsi" controller. Virtio-Scsi controller can be specified as follows:

   <controller type='scsi' index='0' model='virtio-scsi'/>

Notices

If we attach a Virtio-scsi volume to an instance already with a scsi controller other than "virtio-scsi" model, we must first attach a "virtio-scsi" controller with a "index" value different from exists scsi controllers, and then attach the volume and set the "controller" in "address" sub-element of disk config to the correct "index" value of the new attached "virtio-scsi" controller.

Related works

  1. Block-device-mapping-v2 extension has already support attach or boot from a volume which is connected to Instance via scsi bus type and multiple disk types. But cannot generate a virtio-scsi controller.
  2. Libvirt-virtio-scsi-driver BP ([2]) whose milestone target is icehouse-3 is aim to support generate a virtio-scsi controller when using an image with virtio-scsi property. But it seems not to take boot-from-volume and attach-volume into account.

Implementation

API support

Add "virtio-scsi" bus type to block-device-mapping-v2 extension, as follows:

--block-device id=xxx,source=volume,dest=volume,bus=virtio-scsi,type=lun

Boot from Virtio-scsi Passthrough volume

In libvirt driver, append the virtio-scsi controller (will be introduced by libvirt-virtio-scsi-driver BP [1]) if we booting from a volume with "virtio-scsi" bus type.

Attach Virtio-scsi Passthrough volume

  1. If current instance have not a virtio-scsi controller, we first attach a virtio-scsi controller with an unique "index" value.
  2. Set the "controller" of "address" sub-element for the volume's corresponding block-device config to the "index" of virtio-scsi controller

References

[1] https://blueprints.launchpad.net/cinder/+spec/multi-attach-volume [2] https://blueprints.launchpad.net/nova/+spec/libvirt-virtio-scsi-driver