---------- Forwarded message ----------
Date: Wed, 28 Apr 1999 11:54:03 -0700 (PDT)
From: byron feng xiao <byronx@u.washington.edu>
To: cse588-papers@cs
Subject: Analysis and Simulation of a Fair Queueing Algorithm
My first impression on this paper was: too much math derivation to
understand.
Basically, the authors state that congestion control implementations can be
done on either the source or in the routers. The congestion control
implementation done on gateways and routers can be broken down into
Adaptive routing, or queueing. There are two types of queueing: FCFS, and
fair queueing. The authors use the measurements of three quantities:
bandwidth, promptness, and buffer spaces to analyse FCFS and FQ. FCFS
fails in a situation where ill-behaved sources sending out more packets and
congesting the network. On the other hand, FQ uses a round-robin queueing
mechanism, and thus, the ill-behaved source can't get more than its fair
shares. The authors define the meaning of fairness mathmatically, and
state that their sense of users in a network is in the context of a source
destination pairs.
At the end, the users state couple problems with FQ: 1. some
source-destination pairs, such as file server or mail server, require more
than its share of bandwidth. FQ is bad in this environment. The authors
argue that FCFS is also bad in this environment. I disagree. It seems
like FCFS is only bad when there are ill-behaved sources in this
environment, because it takes up more bandwidth from the file server or
mail server. However, in the essence of no ill-behaved sources, FCFS seems
to perform better, because file server/mail server can get their fair
shares, while other applications might be starving. FCFS is good for the
file server/mail server in this environment. It's not fair to other apps,
but FQ isn't fair to file server/mail server either. So in a network where
file server/mail server are important, it's better to use FQ. 2. FQ
requires smart and fast gateways. I do not understand why the authors
think this is a problem.
Overall, a good paper to read, except the math derivation part. Short and
consize is how I feel about this paper.