CS262B Reading Summary
The Impact of DHT Routing Geometry on Resilience
and Proximity
K. Gummadi et al
Summary by Feng Zhou
3/31/04
Strong points of the paper are:
- This paper proposes using underlying geometry as a fundamental
aspect in comparing different DHT systems. Rather than comparing
systems as blackboxes with each other, it proposes analyzing to what
extent the geometry allows or limits the DHT system to do in terms of
routing resilience and exploiting proximity. This provides more insight
than black box performance evaluations.
- The analysis of neighbor selection and route selection
flexibility of various geometries is useful. Basically trees have
flexible neighbor selection but cannot select routes. Bufferfly has
none. Hypercubes have route selection flexibility but no neighbor
selection flexibility. Hybrid of trees and sequential neighbors also
have route selection flexibility for non-optimal routes. Trees have
both.
- Trees have good static resilience because they have great
flexibility in route selection. In contrast, pure trees have no route
selection, thus very bad static resilience. Hybrids improve this
greatly.
- As of path latency (stretch), proximity neighbor selection (PNS)
is shown to be more effective than proximity route selection(PRS) in
general. The reason is in part the fact that using PRS can
possibly lead to more overlay hops, which detriments its benefit.