[ovs-discuss] Questions about Open vSwitch QoS

[Ben Pfaff]

On Tue, Sep 14, 2010 at 9:41 PM, Henrique Rodrigues
<henriquesilvar at gmail.com> wrote:
> Hi,
> I have been studying OVS and I have some questions about its QoS.
> I installed the Open vSwitch 1.1.0pre1 on a Xen machine hosting a single VM.
> I followed the QoS Cookbook entry to limit the VM rate to 1Mbps. Through the
> netperf test, I verified that the VM output rate is indeed 1Mbps.
> Then, back to the Xen host, I noticed OVS had attached an ingress qdisc to
> the VM interface (vif1.0). Does this means that running the commands from
> the cookbook has the same results as running "tc qdisc add dev vif1.0
> ingress" and rate limiting the ingress qdisc? What is the difference between
> these two scenarios?

Setting the ingress_policing_rate and ingress_policing_burst just configures an
ingress qdisc, yes.

> The other question is about whether OVS QoS supports borrowing the excess
> bandwidth (as described in the htb user guide [1]) or not. From the
> observations above, I think the answer is probably no, isn't it?
> If so, is there any other way to have a max-min rate limiting with OVS?

OVS can also configure the HTB qdisc.  Configuration is more complicated but
also more effective.  It is documented as part of the QoS and Queue tables
described here:
       http://openvswitch.org/ovs-vswitchd.conf.db.5.pdf
There is no corresponding "cookbook" entry yet.  (Also, this form of QoS
effectively requires use of an OpenFlow controller at the moment.)