On Wed, 15 Jun 2011 18:43:23 -0700 Lorenzo Colitti <lorenzo@xxxxxxxxxx> wrote: > On Wed, Jun 15, 2011 at 6:21 PM, Mark Andrews <marka@xxxxxxx> wrote: > > > > ... about 80% of the time. > > > > Or 99.999% of the time once you get it setup. The problem isn't 6to4, it's > > *automatic* 6to4. > > > > No, because you often end up being dependent on the whims of BGP and > third-party relays for your return path. If you generalise that statement a bit, "No, because you often end up being dependent on the whims of BGP and third-party routers for your return path." you're describing the trust inherent in the operation of the Internet, both for IPv4 and IPv6. Yes there are some incompetent operators out there, but most of them aren't - they wouldn't keep their jobs otherwise, and the Internet would break more often than it does. I can only remember one instance of 6to4 relay breakage in the last 5 years, and IIRC that was fixed within 24 hours. People seem to be using a very coarse "less than absolute 100% success means total technology failure" to state that 6to4 as a whole is unreliable. Where is the data to back this up? Did Geoff Huston do any more detailed analysis of the 6to4 data, other than measuring success verses failure rates for 6to4 connections? Could I, for example, have warped his 6to4 failure rates during the test if I used all my 2^16 x 2^64 IPv6 6to4 addresses to purposely create excessive failed 6to4 connections apparently from many different hosts? I'm certainly not saying that exists in Geoff's data, rather asking how do we know that it doesn't? I have a vested interest in anycast 6to4 continuing to exist, because it has been as reliable as any other Internet technology I use. I also think it has some latency advantages over configured tunnels because my IPv6 traffic enters the IPv6 Internet where ever the closest 6to4 relay reachable via IPv4 is. That used to be in Europe (around 14000 Km from me) when I first started using 6to4, now it is only around 3000 Km, and as my ISP is going to deploy a 6to4 relay in the near future, will be around 8Km away. These changes have and will continue to be transparent on my end. I'd also get these latency benefits if I was changing my IPv4 Internet points of attachment, such as if I were travelling internationally. Anycast 6to4 also tends to distribute the IPv6 traffic load better across all the 6to4 globally gateways, instead of concentrating it at likely lower number of configured tunnel entry/exit points. Over time, increased IPv6 traffic will become a disincentive to running a configured tunnel entry/exit point because people will continue to use it even if there is a more local configured tunnel or 6to4 relay they could be using. If people use anycast 6to4, then they inherently get lower latency as a closer one becomes available, and 6to4 relay operators will see lower traffic load without intervention as more 6to4 relays are deployed. > Sure, if you have enough control > over routing in a closed network, you can make sure the right relays are > used. But in that case, why not use configured tunnels? > > > > 6to4 historic is throwing the baby out with the bath water. > > > > On this we know we disagree. _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf