On Mon, Jul 17, 2017 at 01:39:13PM +0200, Peter Krempa wrote:
On Mon, Jul 17, 2017 at 07:27:24 -0400, sferdjao(a)redhat.com wrote:
> From: Sahid Orentino Ferdjaoui <sahid.ferdjaoui(a)redhat.com>
>
> In version 2.7.0, QEMU introduced support of busy polling for
> vhost-net [0]. To avoid paraphrasing original authors of that feature
> and the purpose it I prefer to share a pointer [1].
>
> This patch serie exposes throught the NIC driver-specific element a
> new option 'poll_us'. That option is only available with the backend
> driver 'vhost' and that because libvirt automatically fallback to QEMU
> if the driver is not specified where that option is not available.
>
> The option 'poll_us' takes a positive. 0 means that the option
> is not going to be exposed.
We had a similar attempt to do this for disk polling, but that was
rejected since it's not very straightforward for the users to tune this
variable. I think this falls into the same category.
Here's the discussion for iothread polling:
https://www.redhat.com/archives/libvir-list/2017-February/msg01048.html
Yes, thanks for pointing that out. I do have the same objections to this
patch, as for the previous disk polling patch. I just don't think they
are practical for a appliction to use - they're too low level to expose
to sysadmins, and there's no obvious way for an application to pick
the right settings automatically. So I think we're best served by letting
QEMU pick defaults
Regards,
Daniel
--
|:
https://berrange.com -o-
https://www.flickr.com/photos/dberrange :|
|:
https://libvirt.org -o-
https://fstop138.berrange.com :|
|:
https://entangle-photo.org -o-
https://www.instagram.com/dberrange :|