From: Danny Wyatt (danny@cs.washington.edu)
Date: Tue Nov 02 2004 - 19:06:32 PST
Quantifying the Causes of Path Inflation
Neil Spring, Ratul Mahajan, Thomas Anderson
This paper (as its title suggests) is primarily a reporting of empirical
measurements of path inflation on the internet. The authors measure
inflation at 3 different granularities: within an ISP, between two
ISPs, and across the entire Internet (spanning possibly more than 2
ISPs). They also characterize inflation as being caused by either
network topology or routing policy. Within ISPs, they do not find much
inflation due to either topology or policy, suggesting that ISPs
optimize their internal routes to minimize latency. Between pairs of
ISPs they also detect no major inflation, while discerning late-exit
routing policies that show apparent cooperation between ISPs---a strong
indicator that business decisions are not causing sub-optimal routing.
Where they do find significant, preventable inflation is across the
entire Internet. Specifically, they find the most inflation in the
routing policy for interdomain traffic. But, they show that
business-driven routing decisions to not perform significantly worse
than shortest path calculations---which are already mediocre. The
authors suggest that the problem lies in a paucity of information
provided via BGP, and that if BGP were modified to make better route
calculations possible ISPs would embrace them.
I appreciate the existence of empirical studies to either confirm or
contradict intuitive and anecdotal evidence. The methodology in this
paper seems as sound as it could be given the closed nature of ISPs,
ephemerality of internet topology, and large number of uncontrollable
variables. However, since they did collect data over a relatively short
3 day period, it would have been nice to see repeated collections to
expose or smooth transient idiosyncrasies.
This archive was generated by hypermail 2.1.6 : Tue Nov 02 2004 - 19:06:38 PST