On Jun 9, 2011, at 1:42 PM, Lorenzo Colitti wrote: > On Tue, Jun 7, 2011 at 11:20 AM, Keith Moore <mo...@network-heretics.com> > wrote: > Indeed, that is one of its main virtues. 6to4 decouples application > deployment of v6 from network deployment of v6, and helps reduce the "chicken > or egg" problem. > > No, it does not - in fact, it is the opposite. > > Geoff has presented data that shows that anycasted 6to4 as a connectivity > mechanism has a failure rate of the order of 20-30%.
I don't dispute that data. I just disagree with the notion of discouraging 6to4 in its entirety because of the current problems with advertising 6to4 relay routers using anycast addresses. I suspect that the anycast issues will largely be sorted out before this document can have much of an effect. But nevertheless, I don't have a problem with discouraging this use of anycast. I think it was a noble experiment, and we learned something valuable: Don't use anycast to advertise a service that is provided by a wide range of players, at least not without having some fairly clear guidelines about how to monitor them and weed out the broken ones. > We have data that clearly shows that Mac OS 10.6.4, which uses 6to4 by > default, has a ~50x greater failure rate when connecting to dual-stack > servers than Mac OS 10.6.5 - and the only change is to not use 6to4 by > default. Search the list archives for details. Again, I have no problem with implementations disabling 6to4 by default. Especially given the looming threat of LSN, I became convinced that it was the right thing to do. > So the existence of 6to4 is in itself a significant barrier for IPv6 > deployment for server operators and content providers. non sequitur. Existing server operators and content providers can easily provide 6to4 addresses for their servers and content, which will be used in preference to native v6 addresses. > Application developers should develop using manually configured tunnels, not > 6to4. At least they don't have a 20% failure rate. How do you know? How do you even measure the failure rate of manually configured tunnels in the aggregate? I don't think you can monitor that kind of traffic the way you can 6to4, because the traffic patterns are much more constrained. It's been awhile since I used manually configured tunnels (from a well-known tunnel broker). But the one time I did try them, 6to4 worked better overall - lower latency and lower failure rate. Keith
_______________________________________________ Ietf mailing list Ietf@ietf.org https://www.ietf.org/mailman/listinfo/ietf