End-to-End Routing Behavior in the Internet

In this paper the author tried to exploit the behavior of what is actually happening while routing in the large view. As many of the paper reviewed before, when new ideas are proposed, the author tends to ^Óexplain^Ô the behavior of their proposal. While the following paper tried to do some statistics about those ideas, they either use simulations or use measurement that is under dispute or lack of detail. In this paper the author do the research on the observation of the traffic on the network. Since the objective is to see the end-to-end behavior on the network, the author defines the ^Ó virtual link^Ô, that there is a route exist between the two nodes.

Although they give a brief description of how their experiment is conducted, they did not talk about any detail as what kind of measurement they are taking, they only mention ^Óset of measurements^Ô. Although they did put the whole data set on the internet, they should still briefly mention some concept in designing the experiment/ This lack of detail prevent the reader from understanding how their experiment is rooted if not reading the webpage. And this makes it even harder to understand the explanation they provided for some observation. In section 4.3, they argue that although their data is not large enough to cover the whole view, the characteristic that these measurements fall under the same administration can justify their data. Although this might sound true, they should at least provide some statistics to support their view. Since many of the argument follows need to include this argument. And they should provide more concrete for the assumption they made.

Similar problems occur in several other places. When they use the fact that the distribution of recovery times from routing problems is at least bimodal, they should at least provide some experiment result they have in the paper as they did on table 2 instead of just describing the phanomena in a narrative way.

Despite the problem of representing the result from the experiments, this paper demonstrate a comprehensive way of evaluating and giving explanation of the problems on the internet. Some property may be expected when the protocol is first designed, but the actual effect remained unknown. They provide a quantitive approaches to these properties that might potentially reduce the performance of the data transfer. And this analysis can provide other researcher an insight of how some property the protocol may result and have a better foundation of choosing features of protocol. Like the erroneous routing they observed can tell that the assumption of where the packets might go can not depend on luck.

This paper is recommended for reading for it sets up a good standard of how a experiment should be conducted. Its comprehensive observation also shows that conventional way of doing the experiment, simulation can sometimes deviate badly from the actually situaion. Since the actual routing involved too much uncontrolled condition. And this paper also provides a good reference for future researchers the insight of how the physical network works.