Serving DNS using a P2P Lookup Service
-------------------------------------------
Russ Cox et al.

The paper presents a DHT-based DNS implementation. By using a 
p2p distributed hash table (DHash, on top of Chord), the scheme 
inherits the pluses of DHTs, like fault tolerance and load 
balancing. The incentive for this decentralized architecture 
comes from the shortcomings of the current hierarchical DNS 
implementation.

The paper's strength is represented by this new approach of a 
DNS based on a DTH. However, in my opinion, the paper reflects 
a very poor try to build such a decentralized Domain Name System.

One first weakness is the fact that besides the advantages of a 
DHT, there is also the important disadvantage of heterogeneity. 
The Chord DHT assumes all nodes to be homogeneous. In the real 
world, things are far from being all alike. Some servers may be 
more powerful than others, some may be really slow and so on. 
This problem can actually be explained with the load balancing 
graph on page 3. In that graph, 20% of the nodes store less than 
20 records, 20% of the nodes store more than 200 records. In a 
heterogeneous environment, the former 20% may be really powerful 
nodes meanwhile the latter 20% may be really bad servers. How 
does this affect performances?

One other weakness of their system is the lookup cost (logN) which 
may translate in a relatively high number of hops. In fact they 
state this, as they also mention that DDNS would prove a worse 
solution than the existing DNS. One more thing: maybe they should 
have translated RPC. Is it remote procedure call or what is it?