I would leave it as a separate field in the packet_in. We don't have a way of indicating whether a flow has a valid cookie or not, so we would always have to add a NXM_NX_COOKIE field. Also, the match wouldn't be the same as the flow that was originally pushed, since the cookie is part of the Flow Mod message. I think it makes it cleaner to leave it separate, since that's how it will be if we get it added to OpenFlow 1.3. For similar reasons, I hope we don't have to encode the table ID as an NXM field.
--Justin On Dec 29, 2011, at 1:07 PM, Ethan Jackson wrote: > Please refrain from reviewing this patch temporarily. > > Given recent changes to master, do you think it makes sense to encode > the cookie directly in the packet_in structure, or to encode it as > part of the NXM in the packet_in structure using NXM_NX_COOKIE. > Furthermore, I wonder if it makes sense to remove the table_id from > the match structure, knowing that we could encode it with an > NXM_NX_TABLE_ID in the future if someone needs it. Opinions? > > Ethan > > On Wed, Dec 28, 2011 at 22:27, Ben Pfaff <b...@nicira.com> wrote: >> On Wed, Dec 28, 2011 at 06:38:32PM -0800, Reid Price wrote: >>> Perhaps I should file a feature request for json formatted dump-flows >> >> That idea's not half-bad, actually. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
