This thesis addresses the problem of dimensioning ATM networks in a multi-service scenario for the simultaneous support of different types of services: conversational services, retrieval services and point-multipoint services. We consider the problem in a network operator’s point of view, which aims to find the least operational & maintenance cost physical network that can support the expected traffic with the required quality of service. All the dimensioning problems addressed in this work consider switched services where each service is supported by a logical network. Each logical network is composed by a set of constant bit rate Virtual Path Connections (VPCs) with static routing (uni-hour and multi-hour).

The first problem addressed in this thesis is the dimensioning of ATM networks where all kinds of services are supported by logical networks based on end-to-end VPCs. This architecture greatly simplifies Call Admission Control (CAC) procedures. We propose appropriate Integer Programming models for the problem. The proposed models extend previously known models for conversational services in order to cope with assimetrical traffic (required by retrieval and point-multipoint services) and VPC modeling where one of its end nodes is not known in advance (required by retrieval services when there is more than one server node). We propose an heuristic algorithm based on Lagrangean Relaxation with Sub-Gradient Optimization for the resolution of the proposed models in large networks.

However, this architecture leads to low efficiency use of network resources. Moreover, the required large number of VPCs makes it infeasible for large networks. As an alternative, two different logical network architectures are proposed. In the case of conversational services, a 2-layer hierarchical architecture is proposed for the supporting logical networks. This architecture reduces the required number of VPCs and, simultaneously, limits to three the maximum number of VPCs crossed by each service call (an important issue for switched services where call set-up time is a quality of service parameter that must be taken into account). In the case of point-multipoint services, a tree architecture is proposed for the supporting logical networks. Different Integer Programming and Mixed Integer Programming models are proposed for the network dimensioning problem with logical networks following the proposed architectures. The use of a branch-and-bound algorithm to solve the proposed models has demonstrated that the alternative architectures achieve significant gains in the cost of the dimensioned networks when compared with the costs obtained using logical neworks based on end-to-end VPCs.