I'd like to ask for 6to4 relays inside the (IPv6-capable) data centers. My terminology and understanding are probably a little bit off here, but I think I'm close enough. So, there are two types of relays:

* Relays that turn IPv4 traffic to 192.88.99.1 -- or some other IP, if you want to be difficult -- into IPv6 traffic. This type of relay, I don't care about so much. Any Linode user silly enough to still use 6to4 deserves what they get. (On the other hand, it would be a good way for users to get /48s before you roll out native support for them.)

* Relays that turn IPv6 traffic to 2002::/16 into IPv4 traffic (from 192.88.99.1). This is what I care about most. It's easy enough to switch a client (which is to say one's node) from using 6to4 to native IPv6 -- assuming a /116 is good enough for you -- but it's impossible to convince all of *your* clients not to send you traffic from 6to4 IPs, and I'd rather not make their experience even worse than necessary by relying on public relays for my return traffic to them.

Now, last I checked, the nearest 2002::/16 relays were pretty good -- they were all run by HE, and were only a couple ms away, except for Dallas, which used their Chicago relay for some reason. On the other hand, ones run by Linode would be even better. (For example, right now Fremont seems to be suffering from a route loop, according to the traceroute I found.)

I recognize that this is quite unimportant, and I'm not particularly expecting you to implement it. Now that you offer native IPv6 I'm sure you're all quite happy to get as far from 6to4 as you can, and it would only help a tiny fraction of clients. Still, I've been wanting to throw it out there (I hope I didn't post about it 4 months ago and forget...), in case anyone is feeling bored and masochistic.

<3

_________________
Matt Nordhoff (aka Peng on IRC)