Chord DNS Review

The paper introduces a Chord based p2p system for serving DNS. The motivation for doing DNS lookups in a peer-to-peer manner is simply the nice properties of such systems over the conventional hierarchical design, namely fully distribution of records over nodes which results in better load-balancing (not having the concept of root name servers with very high load) and also better fault-tolerance.

The paper shows the "loss" in the trade-off : We will lose efficiency if we use peer-to-peer DNS, yet we can omit all of the bad properties of the conventional DNS. I also like the statistical work done for analyzing the behaviour of Chord-DNS. specially, their trick to extrapolate the latency of Chord based on the hop-number distribution in Jung et. al. 's paper on conventional DNS. However, I can't confirm whether this approximation goes wrong or not, specially when the Jung. distribution is not independent from the kind of DNS that drives this distribution. In other words, we may not have the same distribution when we change into a peer-to-peer system.

The authors explicitly mention that do not want to address the problem of node failure, however, I think node failures and un-noticed departures are (1) very common in p2p systems, because of the lack of control over nodes and (2) they can significantly impact the latency of lookups -- if occured in large-scale and very frequently. So, the performance they report may even fall more in the case of node failures (and arrivals). From the other hand, the paper could focus more on the latencys of lookups after the network arrives in a more stable state, where popular names are widely cached in peers. Finally, it may be possible to add some market rules in peer-to-peer systems to allow each peer to consume the resources proportional to its contribution to the network, what the current p2p systems do. The same method can be applied into p2p DNS.
Received on Thu Nov 10 2005 - 11:07:57 EST