On the Disaggregation of Optical Networks
Citation:
Diaz Montiel, Alan Arnoldo, On the Disaggregation of Optical Networks, Trinity College Dublin.School of Computer Science & Statistics, 2021Download Item:
AlanDiaz_phd_thesis.pdf (PDF) 22.39Mb
Abstract:
Optical network disaggregation is a novel technological paradigm enabling the mitigation of lock- in vendor constraints imposed within legacy systems, by aiming to standardise the control inter- faces embedded in optical equipment to enable the development and deployment of technology- agnostic remote control systems. This has been mainly enabled by the Software-Defined Network- ing (SDN)/Network Function Virtualisation (NFV) paradigms, which have been deployed in recent years to operate on top of optical network systems. However, due to the complexity of disaggreg- ating the control from optical networking equipment, the consolidation of fully software-defined optical networks has been rather slow. A major challenge has been the lack of testing platforms that enhance the evaluation of optical SDN control procedures. In this thesis, we propose an optical network emulation system to enhance the development of optical control plane research, and use this platform to investigate how to build intelligent optical control plane procedures in disaggregated optical networks.
Firstly, we developed a packet-optical network emulation platform, Mininet-Optical, to enhance the development, testing and prototyping of disaggregated, software-defined optical control plane procedures. Our emulation platform is the aggregation of two subsystems: i) an optical network simulation system, to simulate the physical performance of Optical Line Systems (OLSs); ii) a packet network emulation system, Mininet, which is a widely used emulation platform in the area of SDN in the packet-network domain. With our system we are capable of modelling state-of- the-art transport equipment such as Reconfigurable Add/Drop Multiplexers (ROADMs) composed of Wavelength-Selective Switches (WSSs) and Variable Optical Attenuators (VOAs), Single Mode Fibre (SMF) spans, Erbium-Doped Fibre Amplifiers (EDFAs) and Optical Power Monitors (OPMs). Thus, we can model a wide variety of optical network systems and topologies that may be complex and expensive to deploy in physical environments. By extending the internal composition of the Mininet emulator, we are able to abstract the transport equipment in virtual electronic components (i.e., ROADMs from open virtual switches), allowing us to extend the control plane emulation enabled in these. We then integrated Mininet-Optical with the well-known SDN Network Operating Systems (NOS) Ryu and Open Network Operating System (ONOS). Consequently, we were able to evaluate real control plane procedures (e.g., algorithms and systems) in large-scale scenarios.
Secondly, we evaluated the usage of the Ryu controller for building control plane systems and built our own system. Then, we integrated the SDN controller to Mininet-Optical to study the implications of transmission margins on network capacity. And, we evaluated how can these margins be mitigated by using Quality of Transmission Estimation (QoT-E) algorithms based on analytical modelling of the OLS. Moreover, we evaluate the use of active monitoring components to assist the QoT-E algorithms. For this, we propose three QoT-E models considering different types of monitoring capabilities: i) assuming the monitoring of signal power levels and Amplified Spontaneous Emission (ASE) noise; ii) same as i), plus we use the signal power levels to correct the QoT-E prediction inaccuracies in the Nonlinear Interference (NLI) noise occurring at the optical fibre; iii) assuming monitoring capabilities of signal power levels, ASE noise, and NLI noise, such as reference receiver monitors. With these QoT-E models we also evaluated the issues in monitoring placement, focusing on the advantages of retrieving optical signal data at intermediate locations of an optical link.
Thirdly, we looked at the enhancement of QoT-E modules with Machine-Learning (ML) and deep-learning algorithms. We approached this by assessing the ability of ML algorithms to infer the wavelength-dependent operation of optical network components, with focus on the Wavelength- Dependent Gain (WDG) of EDFAs. For this, we propose the usage of the channel load as an input parameter to train the algorithms, in a feature that we label the active wavelength load. We thus used Mininet-Optical to generate large amounts of data to train and test the algorithms that we evaluated. We began our investigation with a thorough evaluation of the Support Vector Machine (SVM) algorithm, and then we also evaluated multiple algorithms, including: K-Nearest Neighbour (KNN), Linear-Support Vector Machine (L-SVM), Radial Basis Function SVM (RBF- SVM), Logistic Regression (LR), Decision Tree (DT), Artificial Neural Network (ANN), Naive Bayes (NB), and Linear Discriminant Analysis (LDA), Random Forest (RF), AdaBoost, and Bagging. We assessed these algorithms in terms of time to train them and F1 score.
Sponsor
Grant Number
CONACYT Mexico
Science Foundation Ireland (SFI)
Author's Homepage:
https://tcdlocalportal.tcd.ie/pls/EnterApex/f?p=800:71:0::::P71_USERNAME:ADIAZMONDescription:
APPROVED
Author: Diaz Montiel, Alan Arnoldo
Advisor:
Ruffini, MarcoPublisher:
Trinity College Dublin. School of Computer Science & Statistics. Discipline of Computer ScienceType of material:
ThesisCollections:
Availability:
Full text availableKeywords:
optical networks, disaggregation, SDNLicences: