Review - The End-to-End Effects of Internet Path Selection

Although the Internet advertises best-effort routing, It is unclear how well
it approaches optimum path selection. More directly, the effects of path
selection policies are not well known for characteristics such as
round-trip-time, loss rate and bandwidth, as viewed by a user. Savage et al.
assert that, at any given time, 30 to 80% of routing paths do not provide
optimal performance.

The authors suggest two hypotheses for the prevalence of superior paths
throughout the Internet. Firstly, they argue that superior alternate paths
are caused by intentionally avoiding parts of the Internet with poor
quality. Secondly, these alternate paths are created by avoiding congestion.
This is an important claim, because both of these cases are fundamental to
the operation of the Internet.

While the intentions of this study are well rooted, and its results quite
believable, the referenced datasets are obviously questionable. Considering
that this paper was funded by the NSF, Cisco, Intel, etc., I image that a
more accurate dataset could have been obtained or generated. Regardless of
weather or not the results in this study are valid, the authors spend
several pages simply attempting to validate their (many) assumptions (of
which there are many more issues that are not covered).

Finally, the degree to which these alternate paths outperform the chosen
path is not well described in the paper. Having many superior selections
might not be significant if each is only marginally better the the chosen
path. Also, I would argue that the Internet is not intended to guarantee the
best performance for all, but rather it ensure that everyone receives good
performance.
Received on Thu Nov 03 2005 - 02:40:43 EST