On Fri, Oct 30, 2015 at 7:00 PM, Tom Herbert <[email protected]> wrote:
> On Fri, Oct 30, 2015 at 6:42 PM, Anoop Ghanwani <[email protected]> > wrote: > > > > > > On Fri, Oct 30, 2015 at 6:40 PM, Tom Herbert <[email protected]> > wrote: > >> > >> > >> > >> Put the VNID into a TLV then you are guaranteed that people will > implement > >> them! > >> > > And don't forget to make the VNID variable length while we're at it!! > > > +1 > Just to be clear I just threw that out in jest (I know I should be more careful about my jokes). That said, I know it's been discussed that 24 bits, which many of the proposals seem to use, may not be enough. I am sympathetic to that, but I'm also sympathetic to ASIC designers being able to get these proposals implemented. I do like the discussion about implementation in Appendix D of the GUE draft. The problem then becomes one of determining what should be supported in the fast path; once we have done that, we effectively have defined a fixed-header solution. And if we don't do that, we risk lack of interoperability as each vendor independently tries to guess what that set should be. Anoop
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
