Review - The Impact of DHT Routing Geometry on Resilience and Proximity

From: Ian Sin <ian.sinkwokwong_REMOVE_THIS_FROM_EMAIL_FIRST_at_utoronto.ca>
Date: Mon, 21 Nov 2005 09:25:35 -0500

This paper provides insight on DHT geometries that are used by the
myriad of proposed implementation of structured P2P systems to provide
high performance, scalability, replication and resilience for large
distributed systems. All of them scale well and perform well, on the
order O(log N), but the authors focus on other aspects of DHTs that are
equally important. More precisely, they focus on the flexibility
achieved by different geometries and what how this flexibility in
routing selection and neighbour selection perform (path lengths, number
of hops, latency) in the face of a highly transient system.

The strength of this paper is that it presents some important results in
a fairly clear manner. The discussion is supported by graphs and the
authors clearly point out the important results in a question/answer
format, which I believe is very powerful. This paper focuses on helping
the reader understand what are some of the important features to keep in
mind in designing future DHTs to achieve high performance, even in a
highly transient system which is reality!

The main weakness of the paper is that it does not reveal how these
graphs were generated. It is still unclear to me as to whether this data
was computed, gathered from simulation or gathered from evaluation of
real-life implementations. I have the impression that these are
simulated results, with some real-traces for latency sections. Based on
this assumption, it is unclear as to how accurate these simulations are
with respect to a real life implementation. On a minor point, even if I
support the use of graphs to explain, some of the graphs were very
unclear (and confusing) like Figure 7, mainly because the lines were too
similar.

Although the authors admit this work can be improved, I believe that
they are going in the right direction. A more in-depth study will be
interesting, although not necessarily useful for something that has no
use yet!
Received on Mon Nov 21 2005 - 09:24:30 EST

This archive was generated by hypermail 2.2.0 : Mon Nov 21 2005 - 10:43:55 EST