On Nov 28, 2012, at 11:01 AM, Chris Wright <chr...@sous-sol.org> wrote: > * Kyle Mestery (kmestery) (kmest...@cisco.com) wrote: >> On Nov 28, 2012, at 10:34 AM, Chris Wright <chr...@sous-sol.org> >> wrote: >>> * Kyle Mestery (kmestery) (kmest...@cisco.com) wrote: >>>> On Nov 27, 2012, at 9:24 PM, Chris Wright <chr...@sous-sol.org> wrote: >>>>> * Kyle Mestery (kmest...@cisco.com) wrote: >>>>>> + >>>>>> + vxlan_port->port = dst_port; >>>>>> + vxlan_port->count++; >>>>>> + hlist_add_head(&vxlan_port->hash_node, >>>>>> + vxlan_hash_bucket(net, dst_port)); >>>>> >>>>> A little unusual to have a hashtable for this. Is this expected to be >>>>> temporary until IANA port is allocated? >>>>> >>>> The idea is to allow destination port configuration on a per-tunnel basis, >>>> so I needed to track this. As Jesse said, maybe a hash table was overkill >>>> here, but that's a design point I think (e.g. I could have done a linked >>>> list instead). >>> >>> Right, but destination port is meant to be a fixed port number. So I >>> assume this is temporary, or are you saying this is a permanent feature? >> >> This is meant to be a permanent feature. After conversations with Jesse, it >> became clear we wanted to default to the OTV port, but allow destination port >> to be configurable by the user. In a prior email, Jesse noted some VXLAN >> deployments using ports other than OTV. Even after talking with Larry, he >> thought >> it was a good idea to make this configurable as well. > > Right. OTV is commonly, but not exclusively, used now. Post draft > we should expect a final port number. But it's a relatively small set > of functionality to maintain to keep the extra flexbility. In which > case, I think the refcounting needs to be re-reviewed as it looked > suspect to me.
Makes sense. I'll eyeball the refcounting as well while making all the other changes you and Jesse suggested. Thanks, Kyle _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
