From: Masaharu Kobashi (mkbsh@cs.washington.edu)
Date: Tue Oct 26 2004 - 08:59:36 PDT
1. Main result of the paper
The paper presents an allocated-capacity framework that provides
different
levels of best-effort service in times of network congestion based on two
types of bandwidth control: sender-based and receiver based.
It claims that the new framework can achieve different throughput with
high assurance, although the claim is based on simulations with simple
setups.
2. Strengths in this paper
The proposed framework requires most of the complexity necessary for the
implementation to the edge of network. It has a great advantage, since
it makes it scalable and flexible.
If this is widely deployed, it has a possibility that it can be a
rational basis for cost allocation which enables the Internet to fund
for future growth.
3. Limitations and suggested improvements
Although the paper claims it is "predictable" by simple simulations,
it seems for the general users on the real Internet it is quite doubtful
that the framework can provide highly predictable results.
Another weakness is it still depends on the changes to the behavior
of routers consistently. So it will require concerted effort throughout
the Internet, which can be hard to achieve in reality.
It is also vulnerable to malicious hosts and routers which changes
the header info to their advantage.
Finally, but most important, there is a big question on how much demand
there is for the proposed discriminating services from the Internet
users.
Without demand the scheme is useless. If "the first class" and "the
economy
class" are provided, it will be hard to price them and more
problematic is
the fact that the value of the first class will decline as the amount of
the use of the first class increases. Eventually, it can happen that
everyone needs to subscribe the first class just to get acceptable speed
in almost any services because of dominant weight of the first class
on the Internet.
4. Relevance today and future
I do not think it is realistic for the above reasons that the proposed
system will be a good solution to the Internet's congestion problems or
to satisfying different types of demands from users. The greatest problem
is the difficulty of pricing and the diminishing value of the first class
as the first class is
oversold.
This archive was generated by hypermail 2.1.6 : Tue Oct 26 2004 - 08:59:36 PDT