CSC 2231 Review of routing geometry

There are several DHT schemes, such as Chord, Pastry, CAN, proposed
recently. This paper analyzes their fundamental design in flexibility, and
compares their simulated performance in terms of resilience and proximity
under PRS and PIS. It claims better flexibility in neighbor selection and
routing selection could lead to superior performance in static resilience,
path latency as well as local convergence.

In the end, the authors draw an interesting conclusion that the ring
geometry achieves the best overall performance among these listed schemes
since it has unsurpassed flexibility in design. We wonder whether this
conclusion can be extended into other metrics, such as load balancing; and
whether this indicates that all of listed competitive DHT schemes, such as
Chord, Pastry, CAN, all belong to the same design category, and thus have
no big difference in the performance of resilience and proximity. We
noticed these DHT schemes have similar limitation in long overlay
look up latency, security flaws, and are all not suitable for very
heterogeneous environment.

According to this paper, routing geometry greatly constrains the routing
design, so it seems hard to propose alternative DHT schemes that achieve
much more effective routing geometry properties than the ring. Maybe
motivations of designing better DHT should come from better security and
load balancing in heterogeneous networks. Another concern is that whether
the DHT is the key issue in a p2p network; for most of multicast
applications, name look up is not the bottleneck since how to building
multicast tree is more critical.

In brief, this paper thoroughly classifies and compares different DHT
schemes in their routing geometry design, and shows us some inherent
difference and similarity in CAN, Pastry and Chord.
Received on Mon Nov 21 2005 - 08:49:55 EST