[ovs-discuss] switch queue size

[Jiaqi Zheng]

Hi All,

Also is there any command or method in OVS to measure congestion ? i.e.,
when the link load is greater than the link capacity, how could I know ?

Best,
Jiaqi

On Wed, Feb 3, 2016 at 12:38 PM, Flavio Leitner <fbl at sysclose.org> wrote:

> On Tue, 2 Feb 2016 15:44:50 -0500
> Jiaqi Zheng <jiaqi369 at gmail.com> wrote:
>
> > According to my understanding, switch should have a input queue for
> > each input port, and output queue for each output port. Is it right ?
>
> When a packet is received by a netdev device, it is identified as part
> of OVS and pushed for data path processing and in the end the actions
> are executed. If the action is to send out in a port, that happens
> immediately.  The device transmitting the packet might queue in the
> qdisc. Look at ovs_vport_output() -> ovs_vport_send() ->
> send() = dev_queue_xmit().  Here the it gets into qdisc [tc(8)].
>
> That queue is outside of OVS and after that the packet hits the wire.
>
> You could tune the NIC increasing the number of receiving buffers
> available [ethtool(8) -g].  Usually NICs can do 4k, but the default is
> 256 buffers.
>
> --
> fbl
>
>
> >
> > On Tue, Feb 2, 2016 at 3:43 PM, Ben Pfaff <blp at ovn.org> wrote:
> >
> > > Which input and output queues?
> > >
> > > On Tue, Feb 02, 2016 at 03:41:48PM -0500, Jiaqi Zheng wrote:
> > > > You mean input queue and output queue? OVS can configure which ?
> > > >
> > > > Best,
> > > > Jiaqi
> > > >
> > > > On Tue, Feb 2, 2016 at 3:39 PM, Ben Pfaff <blp at ovn.org> wrote:
> > > >
> > > > > On Tue, Feb 02, 2016 at 03:24:57PM -0500, Jiaqi Zheng wrote:
> > > > > > Is there any command to set switch queue size ? i.e. buffer
> > > > > > size.
> > > Larger
> > > > > > queue size can accommodate more packets when congestion
> > > > > > occurs.
> > > > >
> > > > > Which queues are you talking about?
> > > > >
> > >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160203/c4135149/attachment-0002.html>