From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Wed Nov 03 2004 - 09:56:16 PST
Review: Quantifying the Causes of Path Inflation - Spring et. al.
This paper investigates the influence of ISP policies and routing
decisions on the path length between two hosts, who can be arbitrarily far
away. The paper also quantifies this influence in the case of intra-domain
routing, routing between peer ISPs under different peering policies, and in
the case of inter-domain policies that are designed for paths that cross
multiple ISPs.
The authors consider the effects of various ISP topologies and
routing policies on latency. The findings seem to indicate that in the
majority of cases, routing policy is the cause of most path inflation
effects. This also brings into light the limitations of BGP in efficiently
designing policy-based logical routing topologies. Overall, I think this is
a great paper. THe first important thing that the authors have contributed
is the attempt at quantitatively reverse-engineering the policy-based
traffic engineering that the ISPs have put in place to suit the commercial
environment, and understand the effects of these decisions on the
performance of the end hosts. The second thing that this paper does is that
it uses a simple setup to effect what is really a very impressive test. Just
using traceroute data to quantify so much speaks of the elegance of the
solution. Of course, the paper does not lay out any solution to the problems
observed, but one assumes that these will form the subject of a future
paper.
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 09:56:18 PST