From: Daniel Lowd (lowd@cs.washington.edu)
Date: Wed Nov 03 2004 - 04:41:20 PST
This paper analyzed, in detail, different ISP routing policies and their
effect on latency. Though forced to reverse-engineer private policies,
the authors nonetheless succeeded in coming up with a great deal of data,
and some interesting analysis.
I found the itra-ISP methodology and discussion most convincing: assigning
weights to edges makes sense, breaking ties using latency also makes
sense, and the results are predictive as well as descriptive.
The peering policy and inter-ISP analyses were more difficult to follow.
In part, it's hard to picture what these different paths look like, and
one can only guess as to the motivations behind these routing decisions.
The San Francisco routing example was very helpful, though, in
illustrating anecdotally the kinds of things that can happen.
The weakest aspects of this paper were the remaining ambiguities and
underemphasized story. The fact that policies seem to be varied, for some
reason, doesn't make a clear point or suggest a course of action. The
conclusion that BGP might have problems could have been supported well,
but it wasn't. I think that this paper got really lost in its graphs and
explanations of them. I found myself wanting to know if ISPs had any real
motivations to change their policies, or if these increased path lengths
were a significant cost, or if things were likely to get worse... the fact
alone that there's some spike at 40ms in some graph doesn't mean much to
me. Why not use online appendices, leaving more room for interpretation?
Even so, the results are worthy of further analysis and discussion by
other researchers. Furthermore, the roughly outlined methodologies may
inspire more, similar work leading to a clearer understanding of what's
really going on. I believe this paper is relevant as a first effort
towards answering these questions, and as a source of data for researchers
studying routing protocols and policies.
-- Daniel
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 04:41:21 PST