[ovs-discuss] queueing discipline

Justin Pettit jpettit at nicira.com
Wed Jul 4 07:23:10 UTC 2012


If you configure it through ovs-vsctl, an appropriate configuration should be pushed down to the kernel.  We don't expose as many knobs as the tc command, so the configuration is much simpler.  You should be able to verify the configuration pushed down by using the tc command to dump classes, qdisc, etc. 

--Justin


On Jul 4, 2012, at 12:11 AM, selen jia wrote:

> Hi Justin,
>  
> Sorry to bother you.
> I am confused because tc manual is having many paarmeters to configure, class , handle id etc.. but in OVS we have just rate or burst.
> so how we will configure parent -child through OVS vsctl?
>  
> regards,
> Selen
> 
> 
>  
> On Wed, Jul 4, 2012 at 12:32 PM, Justin Pettit <jpettit at nicira.com> wrote:
> That's what you're configuring, so OVS is taking care of the tc configuration.
> 
> --Justin
> 
> 
> On Jul 3, 2012, at 11:54 PM, selen jia wrote:
> 
> > Hi,
> >
> > To verify HTB and HFSC on OVS ,if I am creating queue and setting rate through vsctl command then do I need to do some configuration from "tc" also?
> >
> > Regards,
> > Selen
> > On Tue, Jul 3, 2012 at 12:10 PM, Justin Pettit <jpettit at nicira.com> wrote:
> > The former.  If you want to control the rate going into the switch from the VM, you'd need to use a policer.
> >
> > --Justin
> >
> >
> > On Jul 2, 2012, at 11:35 PM, selen jia wrote:
> >
> > > Hi Justin,
> > >
> > > Is that mean to verify HTB and HFSC on VM interfaces , I have to create queues on VM interface with particular bandwidth/rate and have to send traffic from switch to VM and then check rate .
> > > this is ingress to VM, right?
> > >
> > > OR
> > > after creating queue on VM , I will send traffic from VM and will verify the rate of traffic coming out of VM interface on switch?
> > >
> > > Regards,
> > > Selen
> > >
> > > On Tue, Jul 3, 2012 at 11:43 AM, Justin Pettit <jpettit at nicira.com> wrote:
> > > On Jul 2, 2012, at 10:30 PM, selen jia wrote:
> > >
> > > > It means HTB, HFSC and policing all would work on VM interfaces ?
> > >
> > > Yes, they should.  We just leverage the tc's mechanisms in the kernel.
> > >
> > > > Is this implementation opposite to policing because policing act as ingress for switch perspective and egress for VM interface?
> > >
> > > That sounds correct.  Policing is applied on traffic coming into OVS, and shaping (queueing) is applied on traffic going out of OVS.  So, you just have to think about it from the switch's perspective.
> > >
> > > --Justin
> > >
> > >
> > >
> >
> >
> 
> 




More information about the discuss mailing list