On Thu, Mar 21, 2013 at 5:41 PM, Simon Horman <ho...@verge.net.au> wrote: > I think that the way that actions are marshalled by execute_* functions > and then turned into ODP actions using commit_* has worked well for now > and serves to provide a reasonably minimal set of ODP actions. However > it is difficult to make that approach work well if the ethernet type > of the packet is allowed to change, as is the case with MPLS. So it doesn't > surprise me that we are deploying checks multiple locations in order to > ensure the ODP actions are valid.
I agree. I like the fact that it allows us to combine an arbitrarily complex set of userspace actions into a minimal set of kernel actions and would really like to keep the property. However, the current implementation does seem a little at odds with MPLS. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
