On Mon, 17 Jun 2013 18:10:23 +0300 "Michael S. Tsirkin" <m...@redhat.com> wrote:
> On Mon, Jun 17, 2013 at 10:54:39AM -0400, Luiz Capitulino wrote: > > On Mon, 17 Jun 2013 17:42:50 +0300 > > "Michael S. Tsirkin" <m...@redhat.com> wrote: > > > > > > The 1000ms I talked about is *not* what the guest will see. If there are > > > > events pending, the throttle API just queues the event and returns right > > > > away. I'd even _guess_ that this is faster then emitting the event. > > > > > > If the filter is not updated for 1000ms then that is guest visible: > > > it is not getting packets with the new MAC. > > > > Let me understand this better: the filter is going to be updated by > > libvirt, is that correctly? > > Exactly. > > > If this is right, then I can understand where you came from, but how > > can we possibly control how long it will take for libvirt to take > > action? > > We can't, it's a best effort, but that's also true for some NICs. > At least let's not introduce an artificial delay there. Got it. Okay, I think the flag is fine and won't ask for more evidence then. Although I still wonder if we would do alright w/o the flag and w/o using the throttle API...