On Jun 9, 2011, at 4:03 AM, Mark Andrews wrote: In message < 19FB0BB1-9048-476A-A901-67F962A116B1@xxxxxxxxxxxxxxxxxxxx>, Keith M oore writes: On Jun 8, 2011, at 11:35 PM, Mark Andrews wrote:
Have broken 6to4 relays is *good* for the long term health of the
Internet. Applications should cope well with one address of a
multi-homed server being unreachable. Billions of dollars have
been wasted because this has not been seen as a basic requirement
for applications. It really isn't any harder in most cases to do
this right.
Not that I disagree with the idea that applications should be able to
fail over from one address to another, but ... why do you assume that
the server is multihomed?
Yes, that is a assumption which isn't always true but mostly is now. It's definitely true for the content providers complaining that 6to4 is stopping them deploying IPv6.
I get the latter part. It's the "mostly is now" part that I wonder about. Sure, I use the web a lot. But I never think of the Internet as being only, or primarily, about a small number of applications that users on the outside use to get to content in the center. To me it's about this huge variety of applications, that have all kinds of different traffic patterns. The problem with the broken 6to4 relay on an anycast address is that the
application (or user, or site) doesn't get to choose a different relay.
The site can always pick a differnet relay as long as they know the IPv4 address of one. There used to be lists of them. The anycast address is or should be just a convenience function.
It's *very* convenient. The existence of the anycast service means that there can be a simple check box that says "enable 6to4" or that it can be enabled by default when no native connectivity is available. Which ought to be a great thing.
The problems with that appear to be: - Too many people have decided to "help out" without really understanding what is required of a relay router that is advertised via anycast...along with a lack of recommended practices for management of such routers. (Yay that there's still some sense of Internet community. But of course there is a learning curve associated with anything new.)
- Default address selection rules that favor 6to4-to-nativev6 or even 6to4-to-6to4, over IPv4-to-IPv4.
(Then again there are rumors of some transit networks filtering protocol 41...which strikes me as more malicious.) I have suggested that ISP's could advertise 6to4 relay routers to customers
via DHCP draft-andrews-v6ops-6to4-router-option, this can also be used to
turn off 6to4 when it is known not to work (e.g. firewall, behind a NAT) or
there is working IPv6.
But rather than make the transition mechanism work there is this mind set
that 6to4 needs to be killed.
Indeed. And really, I'd be okay with killing it if native v6 were universally available. But that might take another five or ten years.
Keith
|
_______________________________________________
Ietf mailing list
Ietf@xxxxxxxx
https://www.ietf.org/mailman/listinfo/ietf