|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Re: DHCPv6, was: Re: IPv6 Finally gets off the ground
On Sun, Apr 15, 2007 at 12:38:42PM +0200, Iljitsch van Beijnum wrote: > Sure, but that's because with IPv4, there are only three flavors: > > - manual configuration > - PPP > - DHCP Although nobody uses them: - BOOTP - RARP The distinction of DHCP, BOOTP, and RARP is important I think, and it would be good to remember the reasons for that progression, the lessons we learned on the way. If the progression from SLIP or HDLC to PPP also represents a progression in your view as it does in mine, then it is also important to remember. Both of these two progression trees represent the cumulative formulation of knowledge: Users are stupid. Automatic is not just best, it's the only way. > The DHCPv6 servers and clients that I tested two years ago didn't > even support address assignment to hosts. That sounds about right. The interesting events here have been this year or last. > >What DHCP and PPP did do, was to remove all of that, and make ISP > >integration of customer premise something that could "just happen" > >without any handholding or bearded geekery. > > Fortunately, the IETF got things right the sixth time around (?) by > adding the stateless autoconfig to IPv6, so these additional > mechanisms aren't necessary. Forgive me for saying (I do not mean it rudely), that I think this one sentence measures best precisely how far you've missed my point by. It is not enough to observe that the end host has been given an IP address, a prefix is imagined as part of that, and a default gateway. RARP and ICMP router discovery taught us this. It is still not enough to, after several years of thinking this was enough, throw in domain-search and nameserver configuration state. BOOTP taught us this. The main point, is that if you leave "all other host configuration" details up to, well, the host itself, then in practice what you're really doing is leaving it up to the user. Ultimately, it is mandatory that the end-user make a choice in this model, if not about everything, then about "some things". This is intolerable in an ISP environment. Compare it to the current IPv4 network, and you see that no choice is mandatory. You just plug in and go. You might, optionally, over-ride any DHCP or PPP delivered knob, but it is easy to simply return the client to "get everything dynamically" and Just Work (tm). > And exactly how often do people type in the address of their own > system...? I'm thinking more of the 'gamer' demographic, wherein other people type in your IP address. > A problem with the DNS and IPv6 is that unlike IPv4, you can't pre- > populate the DNS so that each host has a valid DNS name as soon as it > receives an address. Manual configuration is problematic for more > than the obvious reasons: host may use temporary IPv6 addresses with > random lower bits to avoid exposing their MAC address. The only > reasonable way to solve this is with dynamic DNS updates. That's an excellent summary. Neither has RTADV supported dyanmic dns updates for years, nor is it likely to in the future. If it does, I would be surprised if it manages to work properly. > This would > be bad except that customers will usually have their own prefix in > IPv6 so this should be solvable security-wise. It may not even involve DDNS, but rather be entirely internalized on the customer's home gateway. I think from everything I have just heard from you, that we could both agree: There have been IPv6 implementations "for years." There has not been IPv6 support until very recently, this year or last depending on how you count. -- David W. Hankins "If you don't do it right the first time, Software Engineer you'll just have to do it again." Internet Systems Consortium, Inc. -- Jack T. Hankins Attachment:
pgp00020.pgp
|