Bringing cognition to multilayer transport networks
- Autores: Alba Pérez Vela
- Directores de la Tesis: Luis Velasco Esteban (dir. tes.), Marc Ruiz Ramírez (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2018
- Idioma: español
- Tribunal Calificador de la Tesis: Ramón Aparicio Pardo (presid.), Gabriel Junyent Giralt (secret.), Ramón Casellas Regi (voc.)
- Programa de doctorado: Programa de Doctorado en Arquitectura de Computadores por la Universidad Politécnica de Catalunya
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Operator’s transport networks are becoming increasingly more complex due to the large number of network layers needed to support the ever increasing number of new services (e.g., video-on-demand, social media, on-line gaming, or video and voice calls) with stringent requirements like very low latency and high throughput, as well as the number of users of those services. As a result, innovative approaches for operating the networks is mandatory in order to fulfill those tight requirements, while reducing costs. The main objective of this PhD thesis is improving network operation by introducing autonomic networking capabilities. To this end, we study algorithms targeting network healthiness by monitoring both, the optical (L0) and the packet (L2) layers. An in depth study concerning centralized vs distributed architectures is carried out for anticipating anomaly or degradation detection before enough to give time to re-optimization algorithms. This will allow to plan the most adequate re-optimization that will end in, e.g., re-routing those affected demands according to their Service Level Agreement (SLAs) aiming at reducing the traffic affected by the detected degradation. This main goal is achieved by the following five specific goals: i) Traffic Anomalies at the packet layer. A score-based anomaly detection method is proposed for improving single Origin-Destination (OD) traffic anomalies detection. In addition, a method is devised to deal with the case of multiple related traffic anomalies triggered by an external event. By anticipating whether other ODs are anomalous after detecting one anomalous OD pair, the number of network reconfigurations, total reconfiguration time, as well as traffic losses are improved. ii) Failure detection and localization/identification at the optical layer based on BER monitoring. BANDO and LUCIDA algorithms are proposed to, first, detect significant BER changes in optical connections, and then, to identify the most probable failure pattern. Devoted to soft failure localization, two techniques for active monitoring during commissioning testing and for passive in-operation monitoring are proposed. iii) Network Reconfiguration. Two reconfiguration algorithms were devised after anomalies at L2 and degradation at L0 are detected. The ODEON optimization problem is proposed to reconfigure the VNT, whereas the SCULPTOR algorithm is proposed to be triggered for demand re-routing after receiving certain BANDO notifications regarding significant BER change. iv) Cognitive Architecture. A monitoring and data analytics (MDA) architecture is devised aiming to reduce the amount of data to be conveyed and to minimize anomaly and degradation detection times. Representative use cases for autonomic networking in multilayer scenarios experimentally validate the distributed MDA architecture presented in this PhD thesis. v) Visualization Techniques. An overwhelming amount of monitoring data is available, yet to be analyzed, it needs to be previously pre-processed. Visualization techniques with specific task-oriented charts are proposed to help operators during failure localization procedures.
