On 08/26/2013 10:10 PM, hzguanqiang@corp.netease.com wrote:
> Hello experts,
[please configure your mailer to wrap long lines]
>
> It seems that blkiotune of kvm domain is set through qemu now,
and We found there are some throttle shaking problem when using this
interface.
>
> As you can do the same thing through blkio cgroup, and this
sounds to show better performance, Can we realize domainSetBlockIoTune
by cgroup just like interface domainSetBlkioParameters ? Or Add a new
interface to support setting disk IOtune by blkio cgroup?
Current libvirt already lets you choose between cgroup blkio vs.
qemu io
throttling. Compare:
http://libvirt.org/formatdomain.html#elementsBlockTuning
describes
<blkiotune> (implemented via cgroups, on a
per-host-block-device basis);
mapped through the virDomainSetBlkioParameters API, or via 'virsh
blkiotune'
http://libvirt.org/formatdomain.html#elementsDisks describes
<iotune>
under <disk> (implemented via qemu, on a per-guest-disk
basis); mapped
through the virDomainSetBlockIoTune API, or via 'virsh
blkdeviotune'
(and yes, we're lousy at using consistently named API)
--
Eric Blake eblake redhat com
+1-919-301-3266