Integer Linear Programming Modeling for Multi-layered Network Optimization Problems

Abstract

With the increasing demand and usage of the Internet, high-speed networks are essential to reduce traffic delay and congestion, providing high-quality services to the end-users. Telecommunication operators constantly bring in new technologies and hardware to enhance the networks, aiming to provide high-bandwidth Internet resources with high quality of service (QoS). The deployment of multiple technologies in telecommunication networks has led to the introduction of layering concept in the network structure. One of the advantages of applying the layering structure is that it can simplify the network design and ease the deployment of various technologies in the networks. Different protocols can be designed in separate layers, thus providing flexibility for the operators to upgrade and deploy new technologies in each layer without affecting the normal operation in other layers. Traffic transmission and switching processes are relatively expensive. Therefore, optimization of network resources becomes a critical concern for the network service providers (NSPs). Due to the tradeoff between total network cost and high-bandwidth services, an efficient optimization model is needed to provide optimal solutions for multi-layered network design problems, aiming to achieve high savings on the total network cost while provisioning sufficient bandwidth to the networks. The key focus of this project is to optimize multi-layered networks using Integer Linear Programming (ILP) model. There are two types of ILP formulations used in the project: the Link-Path ILP Formulation (LPIF) and the Node-Link ILP Formulation (NLIF). In this project, we compare the performance of these two ILP formulations in terms of their cost-effective solutions and the respective optimization time. Throughout the experiments, we observe that NLIF achieves better quality solutions, namely, lower total network cost, than LPIF. However, NLIF requires a longer optimization time as it considers all the routing paths compared to LPIF, which includes only a limited number of routing paths. As the network size increases, we notice that these ILP formulations become progressively inefficient. For large network problems, e.g., NSFNET, we aim to obtain only the near-optimal solutions within reasonable optimization time. From these results, we conclude that ILP formulations are not scalable to solve large-scale multi-layered network problems. However, ILP can provide a near-optimal solution with tight upper and lower bounds, serving as a benchmark for other heuristic algorithms (e.g., MMA) in solving the multi-layered network optimization problems.