In message <4d4b51ea.2030...@brightok.net>, Jack Bates writes: > On 2/3/2011 6:03 PM, Mark Andrews wrote: > > > > The protocol was done in December 2003. Any CPE vendor could have > > added support anytime in the last 7 years. Did we really need to > > specify how to daisy chain PD requests when these vendors have been > > daisy chaining DHCPv4 for various option without any written > > specification? > > > NAT definitely made it easier. The same can't be said for DHCPv6-PD. And > yes, replacing NAT with a protocol that will handle dissemination of > network prefixes deserved having a standards based formula. For CPEs to > work well, there must be expectations of what will happen in a number of > scenarios so that they can deal with it. For example, will the CPE just > hand out /64 networks behind it to other routers? Will it hand out a > prefix one longer than what it received and increment up until it's out > of space? How does this work in the myriad of ways home users connect > things? > > Cheap CPE routers have come a long way over the last decade. They are > probably as close to perfect as you can expect for the price. Now we're > just starting over to go through the pains of trying to automate home > routers. > > > Seriously. CPE vendors could have release IPv6 capable products > > that had a stateful firewall, DHCPv6 with prefix delegation 7 years > > ago. There was *nothing* stopping them except themselves. > > > > People have been retrofitting CPE devices to have this functionality > > for about as long as this. > > Prefix delegation replaces NAT, but there's no standard for how to > divide it up?
Why does there have to be a standard way to divide it up? You fullfill the request if you can or you ask upstream for more, record the result and add a prefix to the routing table pointing at the requesting device. There done. Even with a /48 you are only going to get to 64000 routes which these devices should be able to handle. In practice it will be a lot less. If you don't have a route you send upstream. The ISP doesn't want to have 64000*customers PD leases so it will return a /48. This matches what's done with IPv4 and NATs. This was blindling obvious to me years ago and should have been to any CPE developer. > Sure, people have retrofit it for years. I have myself. > However, even in linux, it's a very manual process and involves deciding > for yourself how you will hand out prefixes behind the front router. > This wasn't a concern with NAT. The most NAT had to worry about was > conflicting addresses on the LAN/WAN (and most, these days, will auto > renumber if necessary). -- Mark Andrews, ISC 1 Seymour St., Dundas Valley, NSW 2117, Australia PHONE: +61 2 9871 4742 INTERNET: ma...@isc.org
