Paper
review: End to End
Routing Behavior in the Internet
Reviewer:
Kevin Hofstra
- How effective is the internet
in routing packets from 2 individual nodes across the infrastructure of
the internet. What accounts for it
reliability, speed, and performance?
- A new evaluation of the performance
of internet packet routing. A study
of how quickly and reliably packets are delivered by investigating
possible shortcomings of internet routing.
It still applies today, especially when many of the services need
to be reliable or even time sensitive.
- The data was gathered using
exponential sampling of traceroute data between
multiple host sites over a varied amount of time. Routing loops account for most of the
undeliverable packets where the propagation of downed links across the
internet account for most of the delayed packets. Different pathologies are being taken
more often as the internet grows and hop count increases, with asymmetric
paths doubling between 94 and 95.
- Critique the main
contribution
- Significance- 2
The article is very effective in showing raw
experimental data of times and routes across the internet, but traceroute is not an effective tool for seeing how
the actual routing is being done in the internet. An end to end tool leaves much to
question about why the data has come out like it has, and this requires a
tool which has intelligence within the network.
- Convincing- 3 The data is gathered very
well, but much of it is inconclusive because the end to end traceroutes don’t tell us much of what is going on
internally.
- From this article we learn
many significant numbers about the reliability and growth of the internet
and what this will do to effective routing. We know that the number
of hops are increasing and the path differentiation increasing, while
the reliability is decreasing. This
leads me to believe we must make changes to our routing algorithms to more
effectively prevent routing loops, and changes to the topologies must be
propagated faster. But I think the
most important question is whether the increase in asymmetry is a good
thing for redundancy and load balancing or whether low path
differentiation and constant paths will lead to more reliable links.