Re: [libvirt] [PATCH 3/4] domain_conf: Add support for iothreads in disk definition
On Wed, Aug 27, 2014 at 05:21:39PM +0100, Stefan Hajnoczi wrote:
On Tue, Aug 26, 2014 at 1:38 AM, John Ferlan
<jferlan(a)redhat.com> wrote:
> + value. Each device must use a unique IOThread and threads will
> + be numbered from 1 to the domain iothreads value.
It is reasonable for multiple devices to share the same IOThread.
For example, if you define just 1 IOThread and bind all virtio-blk
devices to it you have a similar setup as the QEMU (no IOThreads) with
the big exception that the QEMU global mutex is not needed for
virtio-blk I/O request processing. This can already result in a
performance improvement and may make sense where there are many disks
but few host CPUs.
This makes sense. And it's good that John was strict about this
because we will have no problem to relax the checking. One question,
though. Is it also OK to hotplug a disk that is about to share the
iothread with some other, aleady plugged, disk? I assume it is.
Martin
Attachments:
- signature.asc (application/pgp-signature — 819 bytes)