On Planning of Optical Networks and Representation of their Uncertain Input Parameters

Abstract

This paper studies the use of uncertain inputs in the strategic network planning process. To model uncertain planning inputs three essential parameters are needed: the predicted value expressing for instance an expert's view, the uncertainty level indicating the doubt there is about the predicted value, and the confidence parameter denoting the probability that the output parameter was estimated big enough (compared to the actual output). Several planning approaches that handle uncertain variables are distinguished and their strengths and shortcomings are indicated. This allows to indicate the pitfalls in some common planning practices that use a fixed safety margin to handle uncertainty. It is shown that they can lead to incorrect planning decisions, such as underestimation of the impact of the input uncertainty and overdimensioning in case of inaccurately modelled dimensioning problems. Both a theoretic model and simulation results are shown. A real-life planning problem is studied, including forecasting future network traffic from uncertain inputs and dimensioning a network to accommodate an uncertain traffic demand.The network planning process and the changing environmentThe telecommunication landscape has undergone dramatic changes in recent years, especially due to the advances in optical networking. The number of customers grows, as well as the required bandwidth per user, causing the so-called market pull. This effect is very important for Internet traffic [1, 2] and is not expected to stop shortly [3]. Moreover, the operator's cost to be able to accommodate this demand (equipment cost, etc.) is subject to dynamic changes over time as a consequence of the technology push. The latter is apparent from major technological breakthroughs. We are witnessing an important evolution in optical networking technology, moving from point-to-point WDM transmission towards all-optical networks using Optical Path CrossConnects [4]. Finally, the liberalization in most countries has completely changed the telecommunication environment, confronting network operators with fierce competition [5,6]. In this dynamically changing environment, intelligent network planning has even become more important than it used to be [7,8]. Figure 1 gives a graphical overview of the strategic planning process. The overall goal of the process is to provide a network deployment plan optimising the profit of the network operator (trade-off between the expected revenues generated by the customers and the expected costs to realize the network), while taking into account the indirect customer requirements concerning survivability, etc [9][10][11].Especially when planning over a longer time horizon, the uncertainty of the considered planning data becomes important. Forecasting future customer demand is a difficult problem [12][13][14] with an associated unavoidable forecast error,