The End-to-End Effects of Internet Path Selection
-------------------------------------------------
S. Savage et al.

The authors present a study end-to-end Internet performance from the 
users' side. They measure path quality for a high number of datasets 
in terms of RTT, loss rate and bandwidth and for each path between two 
hosts they try to find a better alternative from the paths measured.

I think that the paper strength is represented by the conclusion, which 
is  based on a thorough study. Routing in the Internet isn't optimal 
since it doesn't consider the quality of the paths. 

As a paper weakness, I don't think the authors considered in depth the 
degree by which the better paths have identical links. The section about
Host and AS popularity in alternate paths is somewhat related to this 
although it's not the same focus. Let's consider for example for nodes 
A, B, C, D. We will define path quality in terms of RTT (bandwidth is 
similar). We are interested in A-C and B-C. If A-D + D-C < A-C (< means 
better path) then A-D-C would be an alternate path for A-C. Similarly, 
B-D-C for B-C. The example is for 2 but it can be generalized. How does 
this affect the respective link? If A and B chose the alternate 
path maybe they would make it worse.

Also I think they should have detailed the alternate paths concept in the 
sense that they could have stated that the alternate paths go actually 
through router interfaces for example, "corresponding" to those hosts. 

I don't know how relevant this study is today. At least the loss rate I 
think it is not. I am also not sure about the way they computed the 
bandwidth.