From: Tyler Robison (trobison@cs.washington.edu)
Date: Wed Nov 17 2004 - 07:35:43 PST
This paper describes the history of the service, specifically the
HOSTS.TXT method it was replacing, and the associated scaling problems,
looks at the overall goals and design decisions, and reflects on some of
these choices. The HOSTS.TXT method didn't scale well, and clashed with
the notion of a more distributed system, and so DNS was set up to serve
the same name resolution purposes, and to make the system more
distributed, while being independent of network topology and OS.
While we are given a good account of the DNS service and some of
the history behind it, I would have liked to see more on alternative
systems (apart from HOSTS.TXT), and some indications that this was really
the best option. In what other ways could name resolution be handled?
They do briefly mention existing systems that they considered and
discarded, but we are not given alternatives for the actual choices made
in the system; instead they focus on the choices that were made.
Another problem is the lack of data given; for example, we are
told that the performance was worse than expected, but they provide only
minimal data showing this (in the text), and the analysis feels
incomplete. This is true of the paper in general, and as a result its
fairly difficult to judge the usefulness of DNS.
This archive was generated by hypermail 2.1.6 : Wed Nov 17 2004 - 07:35:44 PST