Abstract—In this paper, a Fuzzy Logic Congestion Detection
(FLCD) algorithm which synergically combines the good
characteristics of traditional Active Queue Management
(AQM) algorithms and fuzzy logic based AQM algorithms is
proposed. The membership functions (MFs) of the FLCD
algorithm are then designed automatically by using a Multi-
objective Particle Swarm Optimization (MOPSO) algorithm in
order to achieve optimal performance on all the major
performance metrics of IP congestion control. The optimized
algorithm is compared with the basic Fuzzy Logic AQM and
the Random Explicit Marking (REM) algorithms. Simulation
results show that the new approach provides high link
utilization whilst maintaining lower jitter and packet loss. The
new approach also exhibits higher fairness and stability
compared to its basic variant and REM.
I. INTRODUCTION
HE Internet has experienced a tremendous growth over
the past two decades and with that growth have come
severe congestion problems. As a result, the Internet is
facing increasing packet loss rates, queuing delays and jitter.
Apart from wasting resources, lost packets can cause
congestion collapse [1] while packet delay and jitter reduce
the quality of service for interactive applications.
Active Queue Management (AQM) denotes a class of
algorithms designed to alleviate problems of congestion
while at the same time ensuring high link utilization and
fairness. Research in this area was inspired by the original
Random Early Detect (RED) proposal [2]. In 1998, the
Internet Engineering Task Force (IETF) recommended the
deployment of RED for congestion prevention. Since this
IETF recommendation, a plethora of AQM algorithms has
been proposed [3], [4].These algorithms have been classified
as heuristic, optimization based and control theoretic. In
spite of all these developments, there has been a slow
deployment of AQM algorithms in commercial circles
because these algorithms perform well only for specific
objectives and under specific scenarios [4]. The
development of an AQM scheme that will achieve
reasonably high performance for all the key AQM objectives
is proving too difficult because most of the objectives are
competing and non-commensurable. For instance, increasing
link utilization comes at the cost of increased packet end-to-
end delay because the buffer is predominantly full. Packet
loss rate also increases due to buffer overflows.
Clement N. Nyirenda and Dawoud S. Dawoud are with the Radio Access
Technologies Centre, School of Electrical , Electronics and Computer
Engineering, University of KwaZulu-Natal, Durban, South Africa(e-mail:
nyirendac@ukzn.ac.za , dawoudd@ukzn.ac.za ).
It has also been shown that, as capacity or delay increases,
traditional AQM schemes eventually become oscillatory and
prone to instability [5]. It has further been pointed out in [6]
that these schemes demonstrate instability with the
introduction of high bandwidth-delay links because they
require a careful configuration of non-intuitive control
parameters. As a result, they are non-robust to dynamic
network changes. They exhibit greater delays than the target
mean queuing delay with a large delay variation, and large
buffer fluctuations, and consequently cannot control the
router queue. It is also stressed that in actual queuing
systems, the mean packet arrival rate and the service rate are
frequently fuzzy i.e. they cannot be expressed in exact terms
[7].Recently, the European Network for Intelligent
Technologies (EUNITE) Roadmap [8] has pointed out that
the application of fuzzy control techniques to the problem of
congestion control in IP-based networks is suitable due to
difficulties in obtaining a precise mathematical model using
conventional mathematical analytical methods. Based on the
fuzzy logic theory [9], several fuzzy logic AQM schemes
[6], [10], [11], [12] have been developed with satisfactory
results compared to the traditional approaches. The general
trend in these schemes is that they use queue length, queue
variation (traffic arrival rate) as inputs to the Fuzzy Logic
Controller. The system output is a probability with which
packets are either marked dropped or marked if Explicit
Congestion Notification (ECN) is enabled. ECN is a means
of explicitly notifying end-hosts of network congestion,
instead of dropping packets. Although the performance of
fuzzy AQM algorithms is generally better than that of
traditional approaches, their major shortcoming lies in the
fact that their control rules and membership functions are
obtained through a manual tuning process which is based on
the designer’s insight. The human factor involved in this
operation makes it difficult for these algorithms to achieve
optimum performance for all the key AQM objectives. The
other problem is that these algorithms are generally designed
with an assumption that the Internet is predominantly
composed of TCP traffic, whose sources respond to
congestion notification signals from routers by reducing
their sending rates. Practically, the situation is not like that
because apart from the non-responsive UDP traffic which
accounts for (22±11) % of Internet traffic [13], the Internet
is nowadays facing a growing list of non-responsive flows
and anomalies such as Denial of Service (DoS) attacks and
routing loops [14]. These flows do not reduce their sending
rates in times of congestion as responsive TCP flows reduce
their rates. Therefore, fairness diminishes exponentially as
Multi-objective Particle Swarm Optimization for Fuzzy Logic Based
Active Queue Management
Clement N. Nyirenda, Student Member, IEEE and Dawoud S. Dawoud
T
0-7803-9489-5/06/$20.00/©2006 IEEE
2006 IEEE International Conference on Fuzzy Systems
Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada
July 16-21, 2006
2231