From: Shobhit Raj Mathur (shobhit@cs.washington.edu)
Date: Tue Nov 02 2004 - 22:56:05 PST
Quantifying the causes of path inflation
========================================
Path inflation in the Internet has been observed for long, but no prior
work has completely explained the effect. This is due to the fact that
most ISPs do not share information about their topology or routing policy.
This paper uses a trace driven study of 65 ISPs to quantify the extent to
which different factors (topology and routing policy within an ISP,
between neighboring links and sequence of ISPs) inflate paths. The results
presented are interesting and contrary to normal expectations.
The authors find that routing within an ISP is almost optimal and does not
cause any path inflation. They identify the main causes of path inflation
as routing policies between neighboring ISPs and inter domain routing. One
of the most striking observations is that there is significant cooperation
between adjacent ISPs in the internet. This is done to avoid non-optimal
routes and for load sharing across links. The other important observation
is that, the main cause for path inflation is the design of BGP (the
current inter-domain routing protocol). BGP does not have sufficient
mechanisms to enable better path selection. At the peering level, BGP does
not provide a convenient way to find an optimal exit routing policy or to
balance load across peering links. At the inter-domain level. BGP does not
provide enough information to enable good AS path selection. As a result,
peering and inter-domain routing turn out to be the primary causes of path
inflation. It is also interesting to learn that a pair of ISPs may not
behave the same way with each other.
The trace duration does not seem to be long enough. Moreover, since the
approach is trace driven it is limited to the set of vantage points and
ISPs chosen. The inter-domain routing and peering policies are inferred
from the observed traces, this is not very convincing.
Overall, the paper presents some results which are contrary to popular
belief. This was the most interesting aspect of the paper. It quantifies
the factors which cause path inflation. Though these numbers may not be
very accurate they give us a useful insight. The paper is well written and
I particularly liked the introduction which summarizes the results and
assumptions in the paper. We can conclude that ISPs should be provided
with better global information (e.g. geographic information of other ISPs)
to reduce path inflation.
This archive was generated by hypermail 2.1.6 : Tue Nov 02 2004 - 22:56:06 PST