qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-devel] Dataplane status for RHEL7


From: Stefan Hajnoczi
Subject: [Qemu-devel] Dataplane status for RHEL7
Date: Wed, 3 Jul 2013 14:26:20 +0200
User-agent: Mutt/1.5.21 (2010-09-15)

Ademar asked for a status update on virtio-blk-data-plane.  Here is my
status, please let me know if you have any questions or concerns.

RHEL7 Features
--------------
For RHEL7 my goal is to deliver the following dataplane features:

1. Image format support
2. QMP 'transaction' support (creating backing file snapshots at runtime)
3. QMP 'block_resize' support
4. I/O throttling support

1-3 have been posted in an RFC series to qemu-devel.  I will send out
patch series at the end of this week for QEMU 1.6.  This will also
include patches for 4.

I'd like to ask IBM to help in the performance evaluation.  We must
compare against the upstream dataplane code, which bypasses the block
layer.  It would be bad if the new dataplane with the block layer is
significantly slower :).

Backport
--------
Since RHEL7 is based on QEMU 1.5.x the backport should be reasonable.  I
don't foresee any particular challenge since the delta against upstream
will be small.

This backport should still happen during the RHEL7 Development Phase
(ending 2013/08/14) since QEMU 1.6 release is 2013/08/15 and the patches
must be upstream before that.

RHEL7.1 Features
----------------
What remains to be done after QEMU 1.6 (RHEL 7.1?):

1. Block jobs
2. NBD exports

  The solution for 1 & 2 is to run the code in the dataplane thread.

3. Hot unplug

  This requires revisiting our crude bdrv_in_use() mechanism, perhaps to
  protect against any remaining commands that would interfere with
  dataplane.

4. Live migration

  We lack something like vhost log for dataplane.  The dataplane thread
  must be able to queue dirty memory areas (accessed by I/O and vring
  processing) so that live migration does not miss pages that were
  modified in the dataplane thread.

Stefan



reply via email to

[Prev in Thread] Current Thread [Next in Thread]