This is an interesting paper that exploits the overlay routing (i.e.
alternative paths) to infer the efficiency of the Internet path selection
based on several traceroute and TCP traces.
It is well known that Internet routing is usually not the path with the
best QoS from the end to end. This paper reveals how inefficient it could
be. Based on their analysis, around 30% alternative paths are better than
default path according to RTT, loss rate.
However, there are several limitations with their methodologies. First,
alternative paths are from overlay routing, instead of Internet routing,
and thus it cannot reflect the delay of real Internet routing. Second,
they do not show how long that a good alternative path lasts. Maybe the
alternative path is just better than the default path for a very short
time. It may be not reasonable for Internet routing to frequently switch
between several short-living good alternative paths. Third, they do not
measure all paths simultaneously which may hurt their results since the
congestion influences all paths concurrently.
Also, their dataset is not sufficient. In particularly, for each date set,
the number of hosts is quite small, less than 40. It reduces the
occurrence of alternative paths.
In addition, it fails to show the fundamental reasons that cause the
inefficiency of Internet routing. According to their comments, it is not
largely due to congestion. But we are not sure about this conclusion.
Received on Thu Nov 03 2005 - 09:43:40 EST
This archive was generated by hypermail 2.2.0 : Thu Nov 03 2005 - 10:05:16 EST