From: Ethan Phelps-Goodman (ethanpg@cs.washington.edu)
Date: Tue Nov 02 2004 - 17:45:10 PST
Quantifying the Causes of Path Inflation
Neil, Ratul and Thomas
The authors carried out an in-depth measurement study to determine the cause
of path inflation in the internet. Their methodology was to collect a large
amount of trace data from 42 locations around the world. This information
was correlated with 65 large ISPs, and geographic locations were extracted
from the DNS names. The amount of data gathered was impressive, and the
experiment seemed well designed. One measure lacking from the data was the
volume of traffic carried between different links, so they can't make any
correlation between the number of inflated paths and the portion of traffic
that experiences inflation.
The examine six potential sources of inflation: topology and routing at the
level of intra-domain, peering, and inter-domain. Of these, inter-domain
routing and peering policy are found to have significant impact on
inflation. Inter-domain routing is a problem because ISPs don't know which
series of ISPs will give the best path, and peering policy is a problem
because ISPs don't know which peering spot in another ISP is the best place
to exchange a packet.
I found it interesting that the competitive nature of the routing policies
didn't seem to negatively impact performance. For example, I would think
that the choice of early-exit vs. late-exit, which is based partly on
economic factors, would impact performance, but this was not the case. In
fact, it seems that if ISPs had the technological means to cooperate more
they would be inclined to. Towards this end, the authors suggest that
protocols publish more information about the make-up of an ISPs network,
possibly via geographic information.
Ethan
This archive was generated by hypermail 2.1.6 : Tue Nov 02 2004 - 17:45:15 PST