From: Jenny Liu (jen@cs.washington.edu)
Date: Mon Oct 25 2004 - 03:18:57 PDT
"Analysis and Simulation of a Fair Queueing Algorithm" establishes a
definition of "fair" and proposes a queueing algorithm that approaches
the ideal according to the definition. It then presents data from
simulations of the fair queueing algorithm and a first come first served
queueing algorithm with different flow control algorithms.
The fair queueing algorithm meets the primary design goal that it sets
out for itself: fair allocation of bandwidth and buffer space. However,
not much more can be said for it.
The design principle of fairness is meant to solve the problem of the
malicious user, but it does not! However "user" is defined, a
malicious person can still get around the unique idenitifcation given by
the router to a user and hog more than his fair share of bandwidth (even
if it may not be going towards any sort of useful work).
Perhaps even worse, the algorithm requires each router to keep track of
ALL users or conversations that route through it, which is unwieldy at
best and perhaps intractable at worst.
Furthermore, When the number of users gets high, the round-robin base
nature of the algorithm can mean that all users have their packets sent
extremely slowly, perhaps to the point where no useful work is being
done by the router.
Though perhaps a step forward from FCFS, the fair queueing algorithm
presented here has only little more to offer.
This archive was generated by hypermail 2.1.6 : Mon Oct 25 2004 - 03:18:49 PDT