On Friday, Feb 14, 2003, at 06:15 Canada/Eastern, Matthew Luckie wrote:
I'm still of the personal view that the only real benefit of V6 is the larger address space, and all the other dragons we fight from day to day (routing scaling, security, traffic engineering, dns, content routing, overlays, and the QoS thing are just the same in V6 as V4).
with IPv6, we get the chance to allocate address ranges in a sane manner unlike what happened with IPv4. CIDR has not solved the routing table size problem.
I think the concern is that even with the best intentions, operational reality will see people punching wholes in each others' aggregates almost immediately so that they can do things like inter-domain traffic engineering and edge multi-homing. Even if elegant solutions are found for those problems, a pure allocation model requires people to renumber their entire networks every time there's an upstream topology change (e.g. they change providers, or their provider changes providers). Certain aspects of renumbering networks with IPv6 are simplified (due to autoconfiguration, for example) but others are just as difficult as with IPv4 (static address configuration, DNS). The registries have never imposed any restrictions on how operators route assigned blocks of numbers on the Internet, and it's difficult to see exactly how they could start doing so. In the absense of a stick *or* a carrot, why would ISPs bother with the pain of renumbering, or impose that burden on their customers? Allocation of IPv6 addresses by the RIRs has already started, and it is being done in almost exactly the same way as IPv4 (financial barrier to entry, no reasonable request denied, no attention given to routing topology).
It was claimed by Ford et al in 1993 that routing tables in core routers could be reduced from 10,000 entries to 200 if addresses were re-allocated according to continental boundaries and service providers.
... which would take a massive coordinated effort, which would no doubt cause considerable instability on the internet, and which would start to be un-done almost immediately once complete since the topology of the Internet is not static.
As for the "no one is using it" argument: someone give me a native IPv6 path out of NZ to the US and Asia. I can't use it if no one will deploy it and give me addresses to use.
APNIC will give you addresses to use, today. All it would take to support v6 on the APE or the WIX would be for you to find at least one other ISP who wanted to play (there are plenty of options for acquiring v6 prefixes for the exchange fabrics themselves). If you're waiting for the big transit providers to support v6 before you use it, you'll be waiting a long time. Why not run it native in your network, run native over exchange points to reach peers, and tunnel out to other v6 islands in the mean time? Joe