From: Jonas Lindberg (jonaslin@kth.se)
Date: Tue Nov 02 2004 - 01:48:07 PST
Review of N.Spring, R. Mahajan, and T. Anderson's "Quantifying the Causes of
Path Inflation"
By: Jonas Lindberg
In this paper Spring et al. discusses possible causes of Internet path
inflation and investigate their impact on today's Internet usage. The
possible causes discussed are: topology and routing policy within ISPs,
between adjacent ISPs and in sequences of ISPs. For this purpose the authors
collected trace data from a number of computers in different countries and
developed methods for analyzing the data.
The results are rather interesting:
- Traffic engineering within ISPs is common, but has a limited
impact on the path inflation.
- Peer ISPs cooperate more than most people think.
- Using early-exit paths often lead to inflated paths.
- Load balancing can lead to sub-optimal paths.
Some of the strengths in this paper are: they authors are the first to
examine routing policy interference for intra-domain traffic engineering and
for ISP peer policies; the results presented are interesting and useful; the
paper opens up for further research. One obvious limitation is the ISPs
unwillingness to share information about their topology and routing policy
which limits the authors to what can be deducted from measurement data. When
it comes to the layout of this paper, it is overall good, but there might be
some room for improvement in the graphs. I thought some of them were a bit
confusing.
I think the most important conclusion in this paper is that the reduction of
path inflation is a technical problem rather than a commercial problem, and
the authors suggest that the best way of doing this would be to improve the
BGP.
This archive was generated by hypermail 2.1.6 : Tue Nov 02 2004 - 10:48:43 PST