From: Craig M Prince (cmprince@cs.washington.edu)
Date: Wed Nov 03 2004 - 06:31:23 PST
Reading Review 11-03-2004
-------------------------
Craig Prince
The paper titled "Quantifying the Causes of Path Inflation" attempts to
explain the interesting observed phenomenon that certain paths in the
internet are sometimes unusually long. The authors' hypothesis is that
these long paths are caused by issues with routing in the network. Namely
the authors identify three places that could be the cause of inflated
paths: intra-domain routing, ISP peering, and inter-domain routing. And in
each of these places, the inflated path could be the result of either the
network topology and/or because of an ISP's routing policy -- leaving a
total of 6 possible causes of path inflation. The paper then
systematically addresses each possible cause. The authors found that it is
often a combination of ISP peering policy and inter-domain routing policy
that is responsible for the path inflation seen. The authors then blame
the lack of information in the BGP protocol from allowing ISPs to
effectively route between each other.
I was impressed by how much information this paper was able to attain
about network topology and routing information from what is essentially a
black-box system. The authors do a good job of explaining their results
and do a thorough analysis of the various causes of path inflation. The
results seem to provide useful insight into the problem of path inflation.
My biggest concern with this paper is that fact that they are trying to
make quanitative claims with imperfect knowledge. What I mean is that at
each step in their method their is the opportunity for noise to be added
to their results. They begin with a internet map that they generate from
traceroute data (and so may be innaccurate), they then use heuristics to
determine different ISP policies and physical topology. All of these
factors make it difficult to determine exactly how much noise there is in
the results.
Overall, the real take-away results from this paper are that even with
incomplete data it is possible to draw useful results about the internet.
Also, another important aspect is that much of the path inflation is a
result of having different ISPs in the internet. This means that taking
into account the commercial nature of the internet now becomes a vital
aspect whenever we design new protocols and systems and must not be
ignored.
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 06:31:24 PST