a.. [BS98] L. Breslau and S. Shenker, Best-Effort versus Reservations: A
Simple Comparative Analysis, SIGCOMM98
b.. [O00] Andrew Odlyzko, Should flat-rate Internet pricing continue?
Computer Networks, 2000

The two papers study the problem of differentiated services over the
Internet infrastructure: [BS98] focuses on the technical motivations for
deciding between best-effort and reservations and provides a carefully
built analytical framework to evaluate the choices. [O00] discusses the
problem of paying differently for different service qualities and
predicts the failure of differentiated services from the economical
point of view. As Janos put it, the two papers are quite complementary:
the first starts from rather inaccurate (simplifying) assumptions,
builds a careful and rigorous analytical model and reaches some
conclusions that are not (cannot be?)  verified. The second paper starts
from perhaps more realistic assumptions, based on historical evidences
and makes predictions (based on market-related assumptions): it looks at
the evolution of older services (mainly telephony) and argues that the
market won't embrace differentiated services for Internet, as it didn't
for telephony, for reasons related to users prefering simplicity.

Points discussed in class on [BS98]
   a. Applications are:
         i. Elastic: increase in bandwidth increases application's
utility/performance.
         ii. Inelastic:
                1. rigid: can't work with bandwidth less than b, can't
use more than b
                2. adaptive: performance grows fast for some middle
values, and very slow for extremes.

           Is this taxonomy exhaustive? E.g.: can it be that too much
bandwidth damages application's performance? We compared this problem
with caching problem - sometimes, a larger cache does make the problem
worse...
We decided that for some scenarios, e.g. when an application decides on
the bandwidth it requires based on the previous network behavior, it
might happen that increasing the bandwidth decreases overall
performance. The probability of this event is likely to be negligible
(maybe interesting point for future research?).

   b. Load distributions: it is notorious that network load is hard to
predict. Why the choice of these particular three distributions? Nobody
seemed convinced they are the only possible ones and nobody seemed to be
sure any of them is closer to reality than the others. However, we
agreed that this is only a framework, one can plug in any load
distribution one sees fit.
    c. Why do we believe their analysis? (Numerical evaluations would
help.
Are they possible/relevant?) We decided that the model of one backbone,
although very simplified, is worth studying and may trigger future
research leading to interesting practical implications. Unfortunately,
no conclusion has been reached on the relevancy of their evaluations.

Points discussed on [O00]:
    a. Do we agree that flat pricing is where the Internet will go?
    b. Do we agree with author's comparisons with other
networks/systems?
Ian mentioned here the book "Victorian Internet", a book that apparently
presents the same sociological patterns during the first telegraph's
days as in Internet today: people got married over the telegraphs, etc.
    c. Why do we believe the author? Lots of data presented, reviews,
...
What is his personal opinion and what is statistical evidence? E.g.,
French vs. US usage of the Internet: prices computed in US $ - relevant
comparison? We agreed that's hard to distinguish between personal
opinion and scientific contribution. However, we also agreed it was an
easy long reading.
    d. One idea for adapting the flat rate to user's usage patterns: let
the user use the service free for a long time, then compute his flat
rate based on his usage. Would this idea work? Users would probably
behave differently for duration of the trial period, but this seems to
be the best pricing model discussed in the paper and in class. Hope is
that users would realize that the trial period affects later pricing and
the difference would not be very significant. (Or if it were, prices
could be adjusted every month based on the history from previous year.)

    e. What are the reasons for the success of flat prices in the past
(for phone companies)? The author claims it's the user's desire for
simplicity. Is that all? We did not agree that the only reason was
simplicity from the user's standpoint, but couldn't quite figure out
what else it could be. (E.g. Long distance calls are not flat! (but are
the local calls?))