[CSC2231] Paper Review: The End-to-End Effects of Internet Path Selection

This paper studies the 5 different datasets to show that the performance
of existing Internet routing is not optimal. The authors find that
alternate paths usually have less drop rate, higher bandwidth and/or
lower round-trip time that help in improving performance.

The strength of the paper is that it recognizes the potential biases of
the trace data and tries to argue why the findings can be generalized. I
believe this is important for any trace-based analysis because the
robustness of the traces directly affect the results of the research.

Although the use of alternate paths seems attractive from the findings
of the paper, I don't believe it is practical at the moment. First, it
requires global knowledge of the Internet routing statistics to find the
best alternate paths used in the paper. It is likely to end up with
another path with average performance if this global knowledge cannot be
obtained. Second, even if the traffic knows its best alternate path, it
requires the cooperation of all routers along the path to deliver this
traffic. Given that existing routers mind their own business when it
comes to routing, a router may route the traffic to a path that is
beneficial to the router itself instead of the traffic. Finally, there
is also the contractual agreement issue as mentioned in the paper. The
alternate paths may pass through ASes that have limitations against
anonymous traffic. Under these considerations, it may be the case that
the existing Internet routing has already made its best effort to route
traffic appropriately.
Received on Thu Nov 03 2005 - 10:05:16 EST