From: Alan L. Liu (aliu@cs.washington.edu)
Date: Wed Nov 03 2004 - 02:13:29 PST
The paper studies the path inflation effect in the Internet. They split
up routing into three cases: intra-domain, peering, and inter-domain.
It was interesting how data was collected without much knowledge of the
actual network topologies involved in the study. Despite this
difficulty, the paper does a good job of providing data and explaining
clearly why it is valid to use. They also do a good job using the data
to support their interpretation of results. Some of the interpretations
were very interesting, such as the conclusion that ISPs are generally
cooperative because they do not tend to exhibit early-exit behavior blindly.
IMO, the crucial aspect of the paper is that it separates the causes of
path inflation into defined cases, each which affect latency for
distinct reasons. This allows the authors to conclude that inter-domain
latency dominates over intra-domain introduced latency. Also,
inter-domain latency is caused by shortest AS-path, not by the
commercial interests that come into play for ISPs.
One limitation of the work is that ISPs using virtual circuit
technologies could not be counted. In class, MPLS was mentioned as an up
and coming solution.
One strange thing about the paper is that it goes to great lengths to
deduce what ISPs' intentions are. I don't understand the commercial
aspects of running a network, so my question is why is it so hard to get
policy information from ISPs?
The paper suggests that improving BGP to include more information would
allow ISPs to make better decisions about how to route .
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 02:13:39 PST