From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Sun Oct 24 2004 - 23:12:43 PDT
Analysis and Simulation of a Fair Queueing Algorithm - Demers, Keshav,
Shenker
This paper discusses the motivations behind, and presents an analysis of the
Fair Queue (FQ) algorithm implemented in the gateways in order to provide
for better flow control in terms of fairness and tolerance to ill-behaved
sources.
The authors note that in order to provide better flow control, the
gateways have to be made more intelligent than treating packets on a FCFS
model. They propose a fair queue scheme that they prove is optimal by
treating fariness as a quantity defined separately in terms of bandwidth
that a source gets as well as in terms of the time required for a pakcet to
be sent out on an outgoing link on a gateway. The FQ algorithm works by
managing different queues for differnt sources based on the amount of time
they spend on an outgoing link relative to the packet size. This provides an
incentive to sources that do not exceed their fair share, thereby making it
harder for malicious sources o deny other users their fair bandwidth. This
also helps prevent congestion as opposed to merely recovering from it when
it occurs, which is a significant step forward.
There are some problems with this method as presented in this paper. Most
important - for differnt classes of service, this method clearly needs many
changes, because the same definition of fairness might not be the
appropriate choice. Clearly decoupling the delay characteristics from
bandwidth is a significant step forward, but some more work is required in
defining differnet classes of service. Secondly, in order to really realize
the potential of this, there would have to be a significant cost in deplying
this algorithm on existing routers. It is also not clear how sources can
make better use of the features provided by an intelligent network core to
optimal advantage, witout overhauling software at the ends. Also, the
overall quality of the paper would have been enhanced had the authors
presented a better picure of FQ with graphs instead of tables, which are
hard to read.
Overall, this paper is definitely a significant stepp forward, and the
authors have made a strong case for FQ.
This archive was generated by hypermail 2.1.6 : Sun Oct 24 2004 - 23:12:44 PDT