Like Ramon, I was puzzled by an API that appears to fail silently if you ask it to match an IP address.
On Nov 7, 2011, at 1:09 PM, Nicholas Bastin wrote: > On Mon, Nov 7, 2011 at 15:03, Josh Smift <j...@bbn.com> wrote: > If they don't specify dl_type, they presumably want both (and any other > ethertype too), just like if they leave any other field blank. > > Doing constant time matching in switch silicon on this type of match is for > all intents and purposes impossible (let alone the fact that the switch isn't > making sure your packet is well-formed). > > NB> - How about tp_src and tp_dst? What ip_proto should we assume that you > NB> are using? > > Does it need to assume that you're using any ip_proto? > > Yes, because (as above), the switch needs to figure out how to parse the > packet data. If you don't tell it ip_proto, how does it know where (in the > packet) to find tp_src and tp_dst in the first place? Sure, you could argue > that NOX (or any other controller) should manufacture the entire set of valid > flowmods if you wildcard a lower layer field, but this would definitely have > to be done in the controller (not the switch), and I would argue is not a > good idea - in a world of finite TCAM space, a controller author should be > explicit about the space they are trying to consume. > > -- > Nick > _______________________________________________ > openflow-discuss mailing list > openflow-discuss@lists.stanford.edu > https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
_______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
