Summary: Review of A Comparison of Overlay Routing and Multihoming Route Control
By Holly Esquivel
The authors present a comparison of overlay routing and multihoming in which they determine
that the relative performance between the two is closer than other studies have shown. They conduct
several experiments in which there conclusions are:
· Given 1-Overlay and 1-multihoming, the overlay outperforms in both RTT and throughput
· Given 1-Overlay and k-multihoming, the multihoming outperforms in both RTT and throughput.
This occurs because the k-multihoming allows much great flexibility in the paths that can be
choosen by the BGP routing protocol in comparison to one overlay
· Given k-Overlays and k-multihoming, the overlays can provide the same or better performance
than the multihoming. If you combine the two to simulate a network that already had
multihoming in place then the results only improved slightly.
The authors also discuss several factors which could have slightly had an effect on their results such as
the amount of traffic on a given day, the key ISPs which were present in the cities they selected and the
fact that their test for path availability could not determine short term failures. Even with all of these
factors though, the biggest point remains that their results indicate that the relative performance of
overlays and multihoming is the same.
One of the pros of this paper is that they discuss the reality of these two routing protocols in
practice. They explain who has to bare the responsibility of making these protocols functional if a
customer wanted to use them. For example, if a customer wanted multihoming most of the burden of