End-to-End Routing Behavior in the Internet
Reviewer: Robert Dugas
This paper deals with the problem of measuring real-world behavior of
routing protocols on the internet.
Various statistics are presented that likelyhood and possible causality
of such issues as 'fluttering', loops and too many hops.
Routing protocols are effective at eliminating loops at the AS level
but less so at finer granularities.
Measuring real world behavior of routing protocols is an important
Route alternation should be examinined as an issue to prevent in
future protocol incarnations.
Many papers have been writing describing routing statics and the
like. Perhaps for its time it was more revolutionary than it now
appears to be.
The method employed to meaure routing statistics was to run
traceroute at various time intervals on a "representative" sample
of webservers then examine the changes or lack thereof.
Regardless of statistics, it is difficult to argue that 37 hosts can
accurately approximate behaviour of the entire internet. Further, tricky
assumptions were made (such as if routing was identical at 2 arbitrarily
spaced sampling points it had not changed in between) to lessen the
validity of some claims. Futhermore, many outage statics are
wildly inaccurate due to the fact that they were not able to be
Highlighted findings were that assymetry is increasing and
should be dealt with. In addition, the presence of loops was greater than
anticipated and longer in duration.