Resumen de Towards cognitive in-operation network planning
Next-generation internet services such as live TV and video on demand require high bandwidth and ultra-low latency. The ever-increasing volume, dynamicity and stringent requirements of these services’ demands are generating new challenges to nowadays telecom networks. To decrease expenses, service-layer content providers are delivering their content near the end users, thus allowing a low latency and tailored content delivery. As a consequence of this, unseen metro and even core traffic dynamicity is arising with changes in the volume and direction of the traffic along the day.
A tremendous effort to efficiently manage networks is currently ongoing towards the realisation of 5G networks. This translates in looking for network architectures supporting dynamic resource allocation, fulfilling strict service requirements and minimising the total cost of ownership (TCO). In this regard, in-operation network planning was recently proven to successfully support various network reconfiguration use cases in prospective scenarios. Nevertheless, additional research to extend in-operation planning capabilities from typical reactive optimization schemes to proactive and predictive schemes based on the analysis of network monitoring data is required.
A hot topic raising increasing attention is cognitive networking, where an elevated knowledge about the network could be obtained as a result of introducing data analytics in the telecom operator’s infrastructure. By using predictive knowledge about the network traffic, in-operation network planning mechanisms could be enhanced to efficiently adapt the network by means of future traffic prediction, thus achieving cognitive in-operation network planning.
In this thesis, we focus on studying mechanisms to enable cognitive in-operation network planning in core networks. In particular, we focus on dynamically reconfiguring virtual network topologies (VNT) at the MPLS layer, covering a number of detailed objectives. First, we start studying mechanisms to allow network traffic flow modelling, from monitoring and data transformation to the estimation of predictive traffic model based on this data. By means of these traffic models, then we tackle a cognitive approach to periodically adapt the core VNT to current and future traffic, using predicted traffic matrices based on origin-destination (OD) predictive models. This optimization approach, named VENTURE, is efficiently solved using dedicated heuristic algorithms and its feasibility is demonstrated in an experimental in-operation network planning environment. Finally, we extend VENTURE to consider core flows dynamicity as a result of metro flows re-routing, which represents a meaningful dynamic traffic scenario. This extension, which entails enhancements to coordinate metro and core network controllers with the aim of allowing fast adaption of core OD traffic models, is evaluated and validated in terms of traffic models accuracy and experimental feasibility.
It shall be mentioned that part of the work reported in this thesis has been done within the framework of European and National projects, namely METRO-HAUL (g.a. no. 761727, 2017-2020) funded by the European Commission, and SYNERGY (TEC2014-59995-R) funded by the Spanish Education and Science Ministry.
