To:
Randy Bush <randy@psg.com>
Cc:
"Matt Crawford" <crawdad@fnal.gov>, ngtrans@sunroof.eng.sun.com, users@ipv6.org, dnsop@cafax.se
From:
"Perry E. Metzger" <perry@piermont.com>
Date:
27 Jan 2001 12:40:43 -0500
In-Reply-To:
Randy Bush's message of "Thu, 25 Jan 2001 11:23:54 +0100"
Sender:
owner-dnsop@cafax.se
Subject:
Re: (ngtrans) Re: IPv6 dns
Randy Bush <randy@psg.com> writes: > >> as fred succintly said, how often does renumbering occur? how often do > >> lookups occur? so, for which should we optimize? > > I can imagine some mobile devices/subnets that are renumbered more > > often than their address is looked up. But, if there's only one nice > > thing about A6, it's that you get to break up the bits wherever you > > want, including nowhere. > > here at ripe, operators and implementors are not impressed with a6, dname, > ... I'm not impressed with A6 -- I'd like it to die. It does not, in practice, solve the renumbering problem. It just causes new trouble in the DNS. The issues in the renumbering problem are database management issues, not protocol issues. I've been saying this for about six years but I don't think I get many people to listen to me outside of folks who've seen things like how MIT Project Athena did machine management. The sad thing is, it isn't just the IETF types who don't understand the lesson -- thousands of sysadmins labor late into the night needlessly because they don't understand how to use relational databases to make their lives simpler. Entire companies, like Tivoli, have been founded and operated on a misunderstanding of the systems management problem. Anyway, no amount of protocol tinkering will help renumbering significantly, any more than any number of bamboo air traffic control towers will cause the planes to land for the cargo cults. The issue isn't protocol support. It is trivial to convey information via a dozen protocols -- we've got plenty to handle all of this. The issue is assuring consistency among hundreds of management databases used in the average network of machines, and that is *not* a protocol problem. Perry