Design and Optimisation of a Novel Composite-Star TDM/WDM Network Architecture: the Petaweb

Because of its tremendous potential, the optical networking is well admitted as a good candidate for the next generation high-capacity communication networks. The attraction of optical technologies grows with the development of the wavelength division multiplexing and agile photonic switching. Photonic switching is advantageous because it significantly reduces power consumption and facilitates switching analogue signals. On the other hand, the heterogeneity of networks such as the Internet, their versatility and also their increasing need in networking optical resources, raise multiple optimisation challenges. The problems are usually difficult to solve and concern the setting up and the engineering of network equipments. This thesis presents the work related to the design and optimisation of a novel high capacity network structure. Such a structure, called the PetaWeb, has a total capacity of several petabits per second (10 15 bit/s). It is composed of an optical agile core that produces high capacity connections between the edge nodes of a network. The PetaWeb is based on an innovative composite-star structure that presents many advantages. In this work we assume the use of the TDM (Time Division Multiplexing) in addition to WDM (Wavelength Division Multiplexing). The thesis aims at proposing methods for the design of a PetaWeb network structure with static traffic and under many conditions. Firstly, we deal with the network model for this peculiar time-slotted wavelength-routed network: switching cores’ structure, switching schemes and routing strategy. The network serves the connection requests through lightpaths of different classes. Routing issues and constraints are analysed in order to interface the PetaWeb to the access network. Given a virtual topology defined by a set of edge nodes and their connection requests, the design problem is to find how to switch lightpaths and where to place switching equipment to plan the network of minimum cost. We call the design problem RFWTAA (Route, Fiber, Wavelength, Time-slot Allocation and Assignment) and we divide it into two subproblems: the resources allocation problem, shortly called RFA (Route and Fiber Allocation), and the resources assignment problem, called WTA (Wavelength and Time-slot Assignment). The first determines the lightpaths routes and the equipments locations; the second proceeds, with linear complexity, to the assignment of wavelengths and time-slots. From the regular composite-star topology given by the RFA solution, a quasi-regular topology is extracted in the beginning of the WTA part of the problem; such quasi-regular topology improves the utilization of the available capacity and the total network cost. Passing through further network upgrades a quasi-regular topology will be modified towards a regular topology. We explain similarities and differences between the RFA problem and the classical location problem. Then we propose two mathematical formulations for the RFA problem resolution: an Integer Linear Programming (ILP) formulation is based on the minimization of a cost function subjected to many constraints; a second formulation considers a dedicated path protection strategy in the network model. The ILP formulations give rise to an extremely hard combinatorial problem; thus, we introduce simplifications to make the problem tractable. The solution of the RFWTAA problem contains the network components geographical location and the lightpaths’ switching schemes. In terms of computational complexity, the resolution time is high for the both mathematical formulations. It is even harder to find an optimal solution when edge nodes are numerous and the traffic matrix is dense. To be able to solve the problem for large instances, we propose the use of an original repeated matching heuristic satisfying the network model constraints. Its results show that the solution is almost the same than the one found by integer linear programming, but with a resolution time significantly reduced. Then we tackle the upgrade problem. We furnish an ILP formulation for the upgrade of an existing PetaWeb network having a regular topology or a quasi-regular topology. You can so appreciate how the idle capacity left available by previous optimisations can be exploited augmenting the network utilization. Finally we compare the results with the case in which the TDM is not used; the benefits in the network cost are important, between the 10% and the 15%.