Modeling, Design and Optimization of Multi-layer Telecommunications Networks


Student thesis: Doctoral Thesis

View graph of relations


  • Yu PENG

Related Research Unit(s)


Awarding Institution
Award date4 Mar 2016


Traffic carried by telecommunication networks has experienced explosive growth regarding both volume and dynamics due to the increasing usage of the Internet all around the world. Emerging new applications and new technologies lead telecommunication networks to a heterogeneous multi-layer multi-technology structure over which multiple services can be provided. For many decades, the design and analysis of telecommunication networks were based on tools developed for the telephone systems, namely, the Erlang formulae. Given the size, complexity and evolving pace of today’s telecommunication networks as well as growing concerns about energy consumption, the original Erlang formulae are no longer adequate. Thus, there is a need for new scalable methodologies to achieve cost effective network designs and operations.
The focus of this thesis is modeling, design and performance analysis of multi-layer and multi-technology networks. The primary objective is to find robust and cost-effective design and dimensioning methodologies of multi-layer communications networks. In this thesis, we propose a near-optimal analytical solution to multi-layer network design problem, which applies to an arbitrary number of layered transport technologies, and realistically sized networks with stochastic traffic models. This solution decides virtual network topologies, link capacity assignments with near optimal choices of routing paths and transport technologies so that total network installation and operation cost are minimized. The key novelty of this solution is the inclusion of stochastic traffic models in a multi-layer network optimization. The incorporation of layering and PPBP traffic introduces a complexity, which cannot readily be managed in the framework of the traditional network optimization methodologies. Fortunately, a radical simplification ensues by adopting the shortest path routing as the overall design philosophy. The optimization algorithm then becomes an iterative application of the shortest path routing together with link-by-link dimensioning, in all layers. There are interesting findings presented in this thesis, which are observed from the empirical results of our algorithm on a range of networks including the 100-node “CORONET” network. Furthermore, a new ILP solution is proposed for the multi-layer network design problem as a benchmark for our heuristic algorithm.
Cost model is another fundamental element in multi-layer network design problems. A correct and detailed accounting of cost model is also the foundation for networking research in the areas of network optimization, network evolution, network design algorithms, network performance evaluation and network management. In this thesis, we propose an idea of using the principle similar to double entry bookkeeping in general accountancy to validate network cost models and thereby the implementation of network design optimization. The key validation technique is to ensure that total network cost calculated from traffic is the same as total network cost calculated by summing the cost of devices and equipment. We show that this principle theoretically justifies pricing based optimization. We also demonstrate how double entry bookkeeping ensures correct results and how optimization fails because of invalid cost, by using the heuristic we proposed as an example. Then we further extend the cost model study to cost error sensitivity analysis of the shortest path routing algorithm. We show that shortest path routing algorithm is an efficient design principle for multi-layer networks, but it can lead to bad designs if the cost models are inaccurate.