On Fri, Aug 9, 2019, 11:53 AM Geoff Huston <[email protected]> wrote: > > > > On 9 Aug 2019, at 10:27 am, Bob Hinden <[email protected]> wrote: > > > > Tom, > > > >> On Aug 9, 2019, at 7:47 AM, Tom Herbert <[email protected]> wrote: > >> > >> > >> As the document highlights the problems of fragmentation are caused by > >> nonconformant middlebox implementations. There is nothing inherently > >> wrong with the fragmentation and end hosts don't a problem with it. > >> IMO, this is just one example of (some) middleboxes arbitrarily > >> breaking end to end protocols. I am hopeful that document will also > >> trigger work to start fixing fix broken middlebox implementations. > >> > > > > I certainly hope so. It’s certainly a big problem for the IETF if any > new protocol is constrained by the worst middle box implementation. > > > The picture gained from broad scale measurements points to a current > reality that removes the conditionality from your observation Bob. > If the application has to work robustly in today’s network then you > need to design defensively and not push into areas where actual network > behaviours deviate from some ideal. I can be as sentimental as anyone > and decry where we have got to as being far from where we had wanted to > be, but I’m also sufficiently pessimistic to believe that there is > insufficient collective impetus to reverse this situation. >
Geoff, The broad measurements are almost always a limited viewpoint taken at point in time. As discussed several times in regards to fragmentation, it's not broken for everyone all the time. It is being used productively in some contexts. The same will be true for other protocols (EH, transport protocols other than TCP) that one may think should be deprecated because they're not "usable" in the Internet. This is precisely why there was a lot of pushback on this draft which originally had much stronger language that would have effectively deprecated fragmentation for everyone, which in turn would have officially validated nonconformant implementations. Tom > ymmv. > >
_______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
