From: Tyler Robison (trobison@cs.washington.edu)
Date: Wed Nov 03 2004 - 00:06:41 PST
This paper investigates 'path inflation', in which the path from
source to destination is much longer than it should be, and what
combination of events and policies bring it about. There has been much
speculation as to why path inflation occurs (such as neighboring ISPs that
refuse to get along), and this paper tries to test the system and get a
stronger understanding of how/why path inflation occurs. As such, the
paper is mostly concerned with collecting data and interpreting this data,
and since ISPs do not generally release much information about their
design, many of the tests are based around figuring out such information.
To summarize the conclusions reached, travel within an ISP did not result
in much path inflation, nor did the topology at the borders between ISPs.
The actual policy implemented for when to moving from ISP to another,
however, appears to play a significant role in causing inflated paths, as
does the policy used to travel across multiple ISPs; these two are the
largest contributors.
The methodology in the paper seems pretty solid; the conclusions
are based on large amounts of data obtained by using methods already
studied and established, and the data itself is well presented.
One nice aspect of this paper is that in addition to the large
amount of data provided, some suggestions are made as to how to fix some
of the problems. But at the same time, many such changes would require
research and testing before being used, and so while interesting, its hard
to say whether these suggestions would actually be helpful, though this is
understandable given that this was not the purpose of the paper. There is
certainly room here for future work, and the paper provides a good
foundation from which to consider the problem.
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 00:06:42 PST