[ovs-discuss] Flushing queues at the OVS

[Ben Pfaff]

On Wed, Jan 18, 2017 at 10:09:33AM +0100, Jorge Baranda wrote:
> I am working in a ring topology with 4 elements, each one running an
> instance of ovs and controlled by an SDN controller in a reactive way.
> As I am having redundancy, I am trying to measure restoration capabilities
> between possible paths after a link down event and link up event.
> When doing iperf tests, the problem comes after the link up event (my
> sequence of actions is: flow stablishment on path1, link down, flow
> stablishment on path2, link up, turn back to path1).
> The transmission stops for a long time and I imagine it could be a matter of
> buffering, which is more critical when testing TCP traffic. I imagine that
> packets get trapped in the switch off interface,
> so when, the interface is up again they are transmitted, but it fact, they
> are not necessary because they should have sent previously on the other
> interface, confusing the TCP receiver and slowing down the process.
> Here comes my question, is it possible to, once the interface goes down,
> flush the packets in the ovs queues? Maybe this is happening also at the NIC
> level? So, I should check both points of buffering, isn't it? Could you
> indicate me how can I also erase the queues at the NIC level?

OVS doesn't queue packets itself, so something else is queuing them,
possibly the device driver.  I don't know how to find or flush those
queues.