From: Jenny Liu (jen@cs.washington.edu)
Date: Wed Nov 03 2004 - 07:33:31 PST
"Quantifying the Causes of Path Inflation" presents the results of an
empirical study that identified classes of ISPs and performed traces to
infer the strategies used for intra-domain routing, peering, and
inter-domain routing. At each of the above levels, the paper examined
the effects of topology and policy on path inflation separately.
The study isolates different possible causes of path inflation in a
systematic way and treats each possibility separately. It also compares
results across continents to see if things are done differently on
different continents and finds a resouding no. It performs traces from
different vantage points across the globe for a better picture of what
is going on. Interestingly, the study points out that early-exit and
late-exit are almost the same thing interms of latency, since one path
is usually the reverse of the other path.
Path inflation certainly decreases global efficiency, but are the
effects of path inflation really so bad? The paper doesn't present the
case for why path inflation is a serious problem in today's Internet.
The worst latency caused by path inflation (due to all factors) at the
95th percentile is only on the order of 60ms, while the mean and median
path inflation is much much lower.
Nevertheless, the study makes clear a good jumping off point for further
work to decrease global path inflation.
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 07:33:29 PST