Machine Learning for Regenerator Placement Based on the Features of the Optical Network

Abstract

With network traffic projected to increase drastically over the new few years, Elastic Optical Networks (EONs) have been brought in to be the successor of the currently used optical technologies. Many factors must be taken into consideration when deploying EONs for wide-scale use. One of which is the overall network's resource allocation. A simple, uniform distribution of regenerators is too inefficient as different locations have different regenerator requirements based on the amount of network traffic they receive. On the other hand, increasing the number of installed regenerators after initial deployment will incur a substantial cost. The ideal scenario is to accurately predict the number of regenerators that each location will need. One way to provide accurate predictions for regenerator allocation is through the use of machine learning. In order to maximize the accuracy of the prediction provided by the machine learning algorithm, it must be supplied with quality input training data. In this paper, we examine the impact that different network features can have on prediction results. We then propose a list of network features that hold significant impact in regards to predicting regenerator allocation accurately.