On Thu, Dec 11, 2014 at 09:46:19AM -0800, Ben Pfaff wrote: > On Thu, Dec 11, 2014 at 09:44:57AM -0800, Ben Pfaff wrote: > > On Wed, Nov 19, 2014 at 09:44:57AM +0900, Simon Horman wrote: > > > Only the default existing behaviour is translated. > > > All other methods are rejected for now. > > > > > > NMX selection method > > > Signed-off-by: Simon Horman <simon.hor...@netronome.com> > > > > I understand why a string is the wire format for the selection method. > > I wonder, however, whether the wire format should be translated to some > > internal representation (such as an enum) at the time we parse the > > action into an ofpact. > > Excuse me, I mean, when the group_mod creates the group.
The idea behind using a string was that it would allow an end-to-end agreement between the controller and the layer that implements the selection method (or not). In Open vSwitch group selection is implemented in ovs-vswitchd and in that context I think your suggestion makes sense. But I do envisage that at some point selection could be offloaded somehow, and at that point it would be the offload implementation that knows which selection methods are implemented. In such a scenario it may or may not be the case that ovs-vswtichd is aware of what the offload supports. Overall my current feeling is that I that implementing your idea does make sense now. And that perhaps it would be good to do so and revisit the problem in the context of offloads if/when it arises in that context. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
