[ovs-discuss] Traffic Shaping in OVS

Ben Pfaff blp at ovn.org
Wed Dec 12 19:32:54 UTC 2018


Next, I'd suggest using ofproto/trace to figure out the path of the
packets through OVS.  It's documented in ovs-vswitchd(8).

On Wed, Dec 12, 2018 at 11:19:52AM -0800, Rallapalli Jagannath wrote:
> Hi Ben,
>  Thanks a lot for your reply. I added min-rate to queue config. Still the
> traffic is going only to the default queue. Do I need to check anything
> else?
> 
> ovs-vsctl set port br1 qos=@newqos1 -- --id=@newqos1 create qos
> type=linux-htb other-config:max-rate=1000000 queues:4=@vn4queue --
> --id=@vn4queue create queue other-config:max-rate=20000
> other-config:min-rate=2000
> 
> :~$ sudo ovs-ofctl queue-stats br1
> 
> OFPST_QUEUE reply (xid=0x2): 2 queues
> 
>   port LOCAL queue 0: bytes=272161446, pkts=193619, errors=0, duration=?
> 
>   port LOCAL queue 4: bytes=0, pkts=0, errors=0, duration=?
> 
> 
> Thanks,
> -Jagan.
> 
> Thanks,
> -Jagan.
> 
> On Wed, Dec 12, 2018 at 7:56 AM Ben Pfaff <blp at ovn.org> wrote:
> 
> > On Tue, Dec 11, 2018 at 11:18:16PM -0800, Rallapalli Jagannath wrote:
> > > Hi,
> > >  I am new to OVS and I have a question on traffic shaping .
> > > Setup:
> > > ---------
> > > I have a bridge br1 with two interfaces br1(LOCAL) and tn1 tunnel
> > > interface. Bridge br1 doesn't have any physical interfaces.
> > > Traffic coming on tn1 goes out of br1 Local interface. It uses iptables
> > > and egresses out of eth0 to the internet.
> > >
> > > I want to apply traffic shaping on some flows from tn1 going out of eth0.
> > > Can someone please tell me how to do this?
> > >
> > > I tried the following and it did not work. Traffic was always going to
> > > default queue.
> > >
> > > // Creating QoS config on br1
> > >
> > > # ovs-vsctl set port br1 qos=@newqos1 -- --id=@newqos1 create qos
> > > type=linux-htb other-config:max-rate=1000000 queues:4=@vn4queue --
> > > --id=@vn4queue create queue other-config:max-rate=20000
> >
> > It looks like you did not define a min-rate for the queue, but min-rate
> > is required.
> >


More information about the discuss mailing list