From: Kate Everitt (kteveritt@yahoo.com)
Date: Wed Nov 03 2004 - 07:55:14 PST
This paper examines the causes of path inflation, a
significant concern in the Internet because it means
that end to end paths can be much longer than
necessary, decreasing the efficiency of the Internet.
At the moment, with overprovisioning being the norm,
this is not as much a concern as it will probably be
in the future, when we may see more contention for
trunk resources. This paper did a good job of
observing policies between peers, including helpful
non-early exit, and load balancing. They also
identified an interesting challenge in work of this
kind: ISPs are unwilling to share topology information
because they consider it proprietary. This work opens
up more work in this field by developing a scheme to
observe network topologies using end to end
techniques.
I felt the most interesting result was that ISPs do
cooperate more than the authors expected. Early exit
paths can cause inflation, but overall inflation
causes trouble for everybody. Traffic engineering has
less of an impact than expected, and load balancing
doesn’t always find the best path. The difference
between networking as an abstract discipline (ie if we
learn in class that routing scheme A is best overall)
and how it actually occurs in practice with a more
competitive organization has an effect on how well
routing schemes work.
This paper was very systematic at looking at path
inflation at three levels, intra-domain, peering, and
intra-domain traffic. However, their data was based on
an inferred about topology, rather than hard data
about how routers are organized. It is possible that
there were other factors or errors that may have
affected the results, but given the information
available they did a very good job of making sure
their results were as robust as possible.
__________________________________
Do you Yahoo!?
Check out the new Yahoo! Front Page.
www.yahoo.com
This archive was generated by hypermail 2.1.6 : Wed Nov 03 2004 - 07:55:20 PST