From: Rosalia Tungaraza (rltungar@u.washington.edu)
Date: Sun Oct 24 2004 - 23:29:44 PDT
This paper is about a modification of Nagle's fair-queuing (FQ)
algorithm used in routers to
separate packets according to the flow they belong to. Moreover, this
algorithm uses a round robin like technique to propagate those packets to
the next link in the network. FQ serves mainly as a congestion control
mechanism, which can be used with various packet-dropping algorithm to
complete that job.
I think one of the successes of this algorithm is that it can both detect
misbehaving users (hosts
who are trying to monopolize the buffers present within a given router and
bottleneck links) and punish/discourage them when they do so. It does that
by offering a separate queue for every flow in the network. Hence, any
flow that sends more packets than what it is supposed to within any given
time period will have its packets dropped. That particular flow will also
be penalized (or charged).
One thing that I would have liked the authors to discuss more is how
their FQ
could be used to deal with issues of Quality of Service. Even though, they
pointed to this issue towards the very end of the paper, I am still
wondering how efficient this algorithm would be if a router was faced with
more than one flow from an intolerant real time application. On the other
hand, QoS in relation to FQ might be a topic that is beyond the scope of
this paper.
As the authors suggest future work in this area could be geared towards
finding out whether one could build FQ gateways that can match the
bandwidth of fibers and if so whether those gateways would be economically
feasible.
This archive was generated by hypermail 2.1.6 : Sun Oct 24 2004 - 23:29:53 PDT