The Design and Implementation of a Next Generation
Name Service for the Internet
--------------------------------------------------
V. Ramasubramanian, E. D. Sirer

The paper presents a DHT-based implementation as a future 
alternative for the current Domain Name System. The authors' 
scheme uses Pastry as the DHT and Beehive as a replication 
framework.

The paper strength is the accurate presentation of the system 
and the incentive for this kind of implementation (the 
shortcomings of the current DNS). The architecture described 
seems really strong. They benefit from the pluses of DHTs, like 
self-organization, scalability, decentralization and in addition, 
they eliminate the high lookup cost by using an efficient 
replication technique. The high cost is translated here in terms 
of high number of hops comparing to the current DNS).

The first weakness comes from the usage of a DHT. In spite of 
their advantages, DHTs have the disadvantage of not performing 
well in heterogeneous environments. Since CoDoNS requires a high
number of servers being deployed all over the world, the 
heterogeneity becomes an issue. One other thing is represented 
by misbehaving peers. If a large number of peers don't respond 
well to queries the system may perform bad.

Another thing is the fact that they state "modifying records on 
the fly, creating synthetic responses" as a negative thing only. 
Content Delivery Networks like Akamai use this DNS Redirection 
mechanism but in a good sense; good from the user's perspective 
who can get the content fast. This is not an issue of abuse or 
security. Furthermore, in their scheme this mechanism wouldn't 
be possible.

Figure 1 is somewhat incorrect. I saw that many people use the 
terms local resolver and local name server interchangeable. In 
fact the local resolver is on the client machine and it is a stub 
resolver in the sense that it relies on a local recursive name 
server to do all the job. Furthermore, that local name server 
(which here is the Resolver) issues only iterative queries to the 
name servers referred (root, tld and so on). Steps 7 and 8 in 
their figure are not correct. There is an exception though; when 
the local name server uses a forwarder. But that forwarder will 
also do the entire job like before and deliver the answer.

All in all, I think the paper is good and the system is strongly 
built.