From: Chandrika N Jayant (chandrika@nyu.edu)
Date: Wed Nov 17 2004 - 01:41:38 PST
“Development of the Domain Name System”
Written by Paul Mockapetris and Kevin Dunlap
Reviewed by Chandrika Jayant
This paper discusses the history of the Domain Name System, a name service for the DARPA internet started in 1983. The previous naming system was a single text file, HOSTS.TXT, which was centralized and very unscalable. Network growth and the move toward IP/TCP Internet from ARPANET led to a trend toward more distributed Internet management. DNS, a variable-depth hierarchical encapsulating name space with typed data at its nodes, created a balance between lean service and a general distributed database. This namespace dealt with the heterogeneity and scalability of the Internet far better than its predecessors.
The authors clearly outline the basic design of DNS: the name space structure, the data format, and how zones and caching help with data distribution. They then discuss the current (1988) status of DNS’s implementation, in particular with regards to root servers and the Berkeley subdomain. They separate the growth of the query rate to DNS (going up) from the rate that is affected by changes in implementation algorithms and timeout tuning(going up or down).
I like the attitude of the paper- it read more like a objective analysis of DNS, not a one-sided proposal which didn’t get the big picture. The authors clearly discuss the successes and shortcomings of DNS, and give good architectural and performance descriptions.I thought it was great that the authors made the point that once most implementers get the level of performance they want out of a system, they will not be motivated to move towards more optimizations so quickly.
A few things that irked me: The authors brought up the use of datagrams as the preferred method to access name servers very late in the paper- it would have made more sense to note this up front. The argument seemed weak about the difficulty making reasonable measurements of DNS performance because of new software releases and gateway changes. I think more tests could have easily been done. I was also worried about security and reliability in general with regards to caching, and would have liked more ideas for future work in this area.
Some notable future work ideas were eliminating up to half of root server traffic with better caching and a less aggressive retransmission algorithms. They could experiment more with different types of queries and see which were really necessary. The need for negative caching seems quite imporant- about ¼ of responses were negative! Integrating access to the naming system into the OS also would be very beneficial.
“Was the DNS a good idea?” Obviously since we still use it, it has passed a good size test of time. The authors note that one cannot assume the form and content of information available is the right way to go- it is just ONE way. Perhaps the whole top-level organizational structure could be rebuilt in a much better way than DNS was. This paper really makes me wonder if we wouldn’t be better off completely redesigning our naming system. Even at Berkeley it was a “major task” to turn the user community to using DNS. How hard would it be now, in 2004, to overhaul our naming system and deploy something new? It seems like it would be a nearly impossible task!
This archive was generated by hypermail 2.1.6 : Wed Nov 17 2004 - 01:41:43 PST