Proceedings of the International Multiconference on ISBN 978-83-60810-14-9
Computer Science and Information Technology, pp. 915 – 920 ISSN 1896-7094
Abstract—Particle Swarm Optimization is applied on an
instance of single and multi criteria network design problem.
The primary goal of this study is to present the efficiency of a
simple hybrid particle swarm optimization algorithm on the
design of a network infrastructure including decisions con-
cerning the locations and sizes of links. A complementary goal
is to also address Quality of Service issues in the design process.
Optimization objectives in this case are the network layout cost
and the average packet delay in the network. Therefore a multi-
objective instance of the hybrid PSO algorithm is applied. The
particular hybrid PSO includes mutation to avoid premature
convergence. For the same reason repulsion/attraction
mechanisms are also applied on the single objective case.
Mutation is passed on to the multi-objective instance of the
algorithm. Obtained results are compared with corresponding
evolutionary approaches.
I. INTRODUCTION
he increasing complexity of the network design prob-
lems calls for advanced optimization techniques. Net-
work design problems where even a single cost function is
optimized are often NP-hard [1]. In addition communication
network design problems are not time critical. Therefore ap-
proaches have been designed to address these problems
based in meta-heuristics such as simulated annealing, taboo
search, evolutionary computing, nature inspired algorithms
or both [2][3]. Concerning evolutionary computing in
telecommunication network design, a comprehensive study is
presented in [4] up to 2005 containing relevant research
study references, where network design problems are classi-
fied in node location problems, topology design, tree design,
routing, restoration, network dimensioning, admission con-
trol and frequency assignment/wavelength allocation. The
optimization techniques employed are mainly variations of
Genetic Algorithms. Additional work on telecommunication
network optimization has followed in the last three years.
T
Real world network design problems normally involve the
simultaneous optimization of multiple and usually partially
contradicting objectives. Therefore more often than not,
there is not a single optimal solution, given the diversity of
the set of objectives, but a set of congruent solutions, known
as Pareto-optimal. The topological design of communication
networks is usually a multi-objective problem involving si-
multaneous optimization of the cost concerning network de-
ployment as well as various performance criteria (e.g. aver-
age delay, throughput) subject to additional constraints (e.g.
reliability, bandwidth). These problem specific objectives
are often opposing; for example a way to reduce average de-
lay in the network is over provisioning; that is to increase
available link capacities which will consequently result in the
increase of the total network deployment cost.
In this paper we will use Particle Swarm Optimization al-
gorithm for the Topological Network Design problem, in-
cluding capacity allocation, considering shortest path rout-
ing. Therefore the target is to design a near optimal network
infrastructure, including decisions concerning the locations
and sizes of links. For that purpose, a hybrid version of the
PSO algorithm will be applied to the real network problem
introduced by Rothlauf [5] and its efficiency will be evalu-
ated against GAs. In addition a bi –criteria communication
network topology problem is considered to address Quality
of Service issues in the design process. For the correspond-
ing delay function, a Poisson traffic model is utilized [1][3]
[7]. This real world application is addressed using multi-ob-
jective PSO. The Pareto front obtained by the MOPSO ap-
plication is compared to the results obtained by a multi-ob-
jective GA (NSGA-II [6]). Relevant work on the subject has
been presented in [1][7] among others using EAs. An alter-
native approach is also proposed in [8] where the relevant
Delay Constrained Least Cost Path problem is addressed,
utilizing the principle of Lagrangian relaxation based aggre-
gated cost, where a PSO and noising metaheuristic are used
for minimizing the modified cost function.
Of crucial importance to the success of the optimization
procedure is the choice of candidate solutions representation.
Especially for evolutionary algorithms a variety of encodings
have been proposed as characteristic vectors, predecessors,
Prufer numbers, link and node biasing, edge sets etc. In [8]
[9] a tree based encoding/decoding scheme, based on
heuristics has been devised for representing the paths as
particles. In the presented work a tree is encoded with the
network random keys (NetKeys) scheme introduced in [10].
978-83-60810-14-9/08/$25.00 © 2008 IEEE 915
Communication Network Design Using Particle Swarm Optimization
C. Papagianni
School of Electrical and Computer
Engineering, National Technical
University of Athens, Iroon
Polytechneioy 9 Str. 15773
Zografou, Athens, Greece
Email: chrisap@telecom.ntua.gr
K. Papadopoulos, C. Pappas,
School of Electrical and Computer
Engineering, National Technical
University of Athens, Iroon
Polytechneioy 9 Str. 15773
Zografou, Athens, Greece
Email: kpapadop@esd.ece.ntua.gr,
chrispap@telecom.ntua.gr
N. D. Tselikas,
D. T. Kaklamani, I. S Venieris
School of Electrical and Computer
Engineering, National Technical
University of Athens, Iroon
Polytechneioy 9 Str. 15773
Zografou, Athens, Greece
Email: ntsel@telecom.ntua.gr,
dkaklam@cc.ece.ntua.gr,
venieris@cs.ntua.gr