From: Danny Wyatt (danny@cs.washington.edu)
Date: Wed Nov 17 2004 - 01:08:46 PST
Development of the Domain Name System
Paul Mockapetris, Kevin Dunlap
This paper provides a historical look at the development of DNS 5 years
after its creation but more than 15 years before the present. Until
this past Monday, I had not known that naming was originally handled by
distributing a single mapping file centrally maintained by SRI.
Obviously, that had to change, although it's amusing that the authors
say DNS might not have, even five years later, fully paid for the cost
of switching to it from HOSTS.TXT. I wonder when the consensus became
that DNS had been worth the switch.
Overall, the paper was pretty lightweight (which I'm thankful for, the
night a homework is due). The few assessments I found most interesting
are: Effect of open, distributed clients on a semi-centralized system:
poor client implementations caused needless strain on the root servers
and are the cause of most of their recommendations for future systems
(e.g., brain dead documentation, query-able version and parameter
settings, possibly even remote updating by the server). "Distributing
authority...does not distribute a corresponding amount of expertise:"
we saw a similar comment in the earlier TCP historical paper. An open
protocol that leaves a lot of decisions up to the end hosts also leaves
a need for sharing extra-standard implementation guidance. Finally,
I've always taken for granted the strong separation between DNS and the
physical (or even logical or political) network. That that was of
uncertain benefit originally was surprising.
This archive was generated by hypermail 2.1.6 : Wed Nov 17 2004 - 01:08:52 PST