On Tue, Feb 14, 2012 at 11:45:17PM -0800, Ethan Jackson wrote: > > + (Before version 1.6, Open vSwitch would drop all packets destined > > for > > + the default queue if no configuration was present.) > > Are you sure this is true? My (admittedly fuzzy) memory tells me that > un-queued traffic effectively reduces the size of the pipe for the > queued traffic. This of, of course, has undesirable results on > min-rates, but isn't as severe as dropping traffic altogether. If we > don't drop unqueued traffic, does forcibly installing a default queue > still make sense? The cost of a misconfiguration is significantly > lower in this case. I'm curious what your thoughts are. I dont' feel > particularly strongly.
In bug #5583 you wrote: > The issue is not that there is no min_bandwidth. The problem is that > there is an OVS QoS object configured in the database which has no > Queues to send traffic through. Thus when traffic destined for eth1 is > forwarded through the QoS object, it has nowhere to put it and simply > drops it. This is a misconfiguration likely caused by the controller and > thus a bug should be filed against them. And then later, when it was reopened for what turned out to be a different reason, you wrote: > There is no default queue so the traffic which isn't hitting the > configured queue has no bandwidth allocated to it. This is expected > behavior in this situation. Looks like a configuration error. Please argue out the situation with yourself and get back to me when you've come to a conclusion. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
