From: Pravin Bhat (pravinb@u.washington.edu)
Date: Wed Nov 17 2004 - 08:55:30 PST
Paper summary: The paper provides a hindsight perspective on the design and
evolution of DARPA's DNS service - why it was absolutely essential at the
time, why it was favored over alternatives, what worked, what didnt
and why, and where should the community go next.
Paper Strengths:
The paper succinctly summarizes the DNS service. It introduces the reader to
the conditions surrounding the internet and the computer industry during the
early 80's which precipitated the need for a distributed naming service over a
centrally maintain repository of names to addresses mapping which was then
locally stored in the HOSTS.txt file. The readers gain a historic view of how
with the introduction of workstations and PCs, the rate of changes to the
HOSTS.txt file and its absolute size grew to the scale where a constantly
updated local database was no longer a feasible solution.
The paper explains the motivations and constraints that shaped the design of
the DNS service -> there was a need to keep it simple to ensure community-wide
acceptance and at the same time the service had to be general enough to justify
the re-engineering. Its amazing to see that the designers didnt solve the
problem at handle by creating a distributed database that simply mapped
names to IP addresses, which is what the HOSTS.txt was most used for. Instead
the designers showed some foresight in designing a general naming service
that mapped names to "values" and the structure of the data values was kept
general enough to support most data types of variable length within reason.
The paper also provides the community with a hindsight view on the evolution
of DNS. It introduces the design concepts that worked, the ones that failed
and should have been avoided, the genuine surprises like the need to
cache negative lookup results and a direction for future research.
Limitations and improvements:
Its amazing to see from so far out into the future how unaware the authors were
of internet's impending success. The authors claim that storing the email
address to mailbox mapping for every user should be a feasible feature for
DNS to support ( as opposed to simply mapping the domains to zones which
would then be accountable for find the user's mailbox) .
Also the paper fails to make any technical contributions and seems more
suited as a book chapter. It simply describes an existing system with a
historic perspective. It would be much harder to publish such a paper today.
Relevance and future work:
The paper is relevant from a historic perspective in the sense that
future designers can learn from previous attempts to design systems that
have had much success. DNS is probably one of the most used services in the
internet today often without the explicit knowledge of the end users.
I'm unsure if some of the observations presented in the paper continue to
hold today given how much the internet changed since then. For example
its not likely that DNS is being used to see if certain networks are now
reachable through IP which at the time the paper was written was a large
enough phenomenon to require the use of caching negative DNS.
Most of the future research directions proposed in the paper have been explored,
so much so that people are now looking into replacing DNS with stronger naming
services like Intentional Naming System.
This archive was generated by hypermail 2.1.6 : Wed Nov 17 2004 - 08:55:35 PST