On Oct 18, 2013, at 6:05 PM, Jian Qiu <sword...@gmail.com> wrote: > I have two questions: > > 1) As megaflow was introduced in ovs 1.11.0, I'm wondering whether megaflow > is transparent to userspace? i.e. is there any specific requirement on flow > configuration so that we can fully utilize the power of megaflow?
The flow table and general configuration defines how much wildcarding can be done in the kernel. The more fields that are looked at, the less wildcarding that can be done. At two extremes, if you just have the default "normal" flow, the wildcarding will be very good (just the L2 headers will be matched), but if you match on the TCP ports, the wildcarding will be worse (L2, L3, and L4 headers will be mateched). Other switch configuration can also affect wildcarding. For example, enabling NetFlow will un-wildcard L2, L3, and L4 headers. > 2) As logs show, we persistently observed "failed to flow_del" messages in > our ovs logs. I'm wondering what's the cause for this warning message? Hmm, that's interesting. Can you provide sample output of "ovs-dpctl dump-flows" so that we can see how much wildcarding is happening in the flows being pushed down to the kernel? In the log snippet you provided, there look to be a ton of errors around one particular flow, which is strange. Would you be willing to run ovs-bugtool and direct mail me the output? I'd like to take a closer look at the complete configuration of your switch and other logs, since I haven't seen that behavior. --Justin _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
