When the current architecture is broken, you have to use what works. Just one example is NAT-traversal. UDP is the only choice. And default-off is a feature.
Dino P.S. Ending thread per Alia’s request to stay on topic of a decision. > On Jul 26, 2016, at 12:46 PM, Joe Touch <[email protected]> wrote: > > Some other points: > > On 7/26/2016 12:22 PM, Dino Farinacci wrote: >> Now, let’s think about this. Why waste 4-bits or a byte for every single >> packet when the UDP port number can be your version number. That UDP port >> number has to be in every single packet anyways. >> >> Why keep the port number the same and change the version number when the >> same cost of product change will occur. To save UDP port numbers? >> >> What if people wanted to filter v1 versus v2, doing it with a UDP port >> number is a simpler and already deployed way to differentiate services. Now >> those middle boxes have to look even deeper into the header? > > Alternately, changing the port number then disables your protocol on all > default-off configurations until it's opened up, which means V1 works > but V2 doesn't. That can be very difficult to debug and track. > > In a nutshell, one of the reasons we're on a good track with the port > number space is our conservative preservation. That includes requiring > new assignments to support in-band versioning. > > The issue is "tragedy of the commons" - the cost of doing this "just for > you" is low, but the cost to the community if everyone has permission > *has already been deemed* too high. > > Finally, while I appreciate some of these considerations, they appear to > be your opinion of the key aspects of IETF protocol design. They are not > consistent with IETF BCPs, nor are they agreed as the best way to > develop protocols that are useful *not only now* but in the future as well. > > Joe _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
