From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Tue Oct 26 2004 - 22:42:19 PDT
This paper proposes a scheme to handle congestion control aiming at
providing QoS services, traffic monitoring and variable fairness.
The idea is to handle bandwidth allocation at the routers to guarantee users
with different requirements certain bandwidth. The scheme uses a single
queue but uses packet marking to discriminate packets that should be dropped
when needed. It also uses random early dropping to handle congestion more
efficiently.
Clearly this scheme can provide better bandwidth allocation than the current
non existing one. It also allows providers to monitor and control bandwidth
among their customers. The customers can also get guarantees that congestion
is more unlikely to make the routers treat them unfairly. The authors
present some simulations to back up the results. Missing from the paper is
some analytical evidence on the performance of the proposed scheme. Although
simulations are useful, analytical results would give more credibility.
Although, the scheme is an improvement on nothing, therefore surely better,
a lot of questions arise for its efficiency. For example, I find the scheme
very dependent on the providers to not overload the network and be fair to
other providers. Additionally, it seems that in order to get any benefit
this scheme would have to be implemented in a large portion of the internet,
something which today might not be feasible.
Overall, the paper tries to address one of the crucial internet design
flaws. Although the scheme is intuitive and simple, I was not convinced on
that its effectiveness is worth the deployment effort.
This archive was generated by hypermail 2.1.6 : Tue Oct 26 2004 - 22:42:12 PDT