Reconfiguration in a Wavelength-Routed Optical Network is a
process of rearranging a virtual topology to meet traffic
demands that change over a period of time. This dissertation
studies a series of reconfigurations corresponding to a series
of changes in traffic demand matrices. A change in a virtual
topology is costly in terms of traffic disruption. However,
without response to this change, the virtual topology would
lose its optimality and might not serve the new traffic demand.
Therefore; the reconfiguration problem is a trade-off between
a performance objective and a cost objective. This research
describes the reconfiguration problem from two perspectives.
First, the reconfiguration problem is a multi-objective
optimization such that a single-objective optimization method
could not be applied. Second, the reconfiguration problem
consists of a series of reconfigurations with corresponding
traffic demands, thus reconfigurations that consider only the
current traffic demand cannot guarantee the optimal average
outcome. Therefore, sequential decision-making is required to
optimize the average outcome from a series of reconfigurations.
Since the reconfiguration objectives are conflicting there
exists a Pareto front or a set of non-dominated solutions in
all objectives. A Multi-Objective Evolutionary Algorithm (MOEA)
is required to search the Pareto front and a decision-making
process will pick one solution in the Pareto front accordingly.
The major contribution of this research is a complete reconfiguration
model applicable to any kind of traffic. It is a stochastic model
consisting of two tasks: a reconfiguration process and a policy.
For each reconfiguration in a series, the reconfiguration process
finds a Pareto front and the policy picks a solution from the Pareto
front to perform a reconfiguration operation. Our research presents
the problem formulation mathematically and the design of the model
is based on realistic SONET/SDH traffic streams. We use a MOEA called
Strength Pareto Evolutionary Algorithm (SPEA) in the reconfiguration
process and use a Markov Decision Process in the policy. A case
study based on simulation experiments is conducted to illustrate the
application and efficiency of the model. It shows that our model
generates a higher average outcome than that of reconfigurations
considering only the current traffic demand.