RT Dissertation/Thesis
T1 Deep learning solutions for next generation slicing-aware mobile networks
A1 Bega, Dario
AB It is now commonly agreed that future 5G Networks will build upon the networkslicing concept. Network slicing is an emerging paradigm in mobile networks thatleverages Network Function Virtualization (NFV) to enable the instantiation of multiplelogically independent copies -named slices- of a same physical network infrastructure.The operator can allocate to each slice dedicated resources and customized functions thatallow meeting the highly heterogeneous and stringent requirements of modern mobileservices. Managing functions and resources under network slicing is a challenging taskthat requires making efficient decisions at all network levels and in real-time, which canbe achieved by integrating Artificial Intelligence (AI) in the network.This thesis investigates the potential of AI for sliced mobile networks. In particularit focuses on resource allocation and orchestration for network slices. This involves twosteps: (i)Admission Control that is responsible to decide which slices can be admittedto the network, and (ii) Network resource orchestration that dynamically allots to theadmitted slices the necessary resources for their operation.Network Slicing will have an impact on the models that sustain the business ecosystemopening the door to new players: the Infrastructure Provider (InP), which is the owner ofthe infrastructure, and the tenants, which may acquire a network slice from the InPto deliver specific service to their customers. In this new context, how to correctlyhandle resource allocation among tenants and how to maximize the monetization ofthe infrastructure become fundamental problems that need to be solved. In this thesiswe address this issue by designing a network slice admission control algorithm that (i)autonomously learns the best acceptance policy while (ii) it ensures that the serviceguarantees provided to tenants are always satisfied. This includes (i) an analytical modelfor the admissibility region of a network slicing-capable 5G Network, (ii) the analysis ofthe system (modeled as a Semi-Markov Decision Process) and the optimization of theinfrastructure provider’s revenue, and (iii) the design of a machine learning algorithmthat can be deployed in practical settings and achieves close to optimal performance.Dynamically orchestrate network resources is both a critical and challenging task inupcoming multi-tenant mobile networks, which requires allocating capacity to individualnetwork slices so as to accommodate future time-varying service demands. Such an anticipatory resource configuration process must be driven by suitable predictors that takeinto account all the sources of monetary cost associated to network capacity orchestration.Legacy models that aim at forecasting traffic demands fail to capture these key economicaspects of network operation. To close this gap in the second part of this thesis, wefirst present DeepCog, a first generation deep neural network architecture inspired byadvances in image processing and trained via a dedicated loss function in order todeal with monetary cost due to overprovisioning or underprovisioning of networkingcapacity. Unlike traditional traffic volume predictors, DeepCog returns a cost-awarecapacity forecast, which can be directly used by operators to take short- and long-termreallocation decisions that maximize their revenues. Extensive performance evaluationswith real-world measurement data collected in a metropolitan-scale operational mobilenetwork demonstrate the effectiveness of our proposed solution, which can reduceresource management costs by over 50% in practical case studies. Then we introduceAZTEC, a second generation data-driven framework that effectively allocates capacity toindividual slices by adopting an original multi-timescale forecasting model. Hinging ona combination of Deep Learning architectures and a traditional optimization algorithm,AZTEC anticipates resource assignments that minimize the comprehensive managementcosts induced by resource overprovisioning, instantiation and reconfiguration, as well asby denied traffic demands. Experiments with real-world mobile data traffic show thatAZTEC dynamically adapts to traffic fluctuations, and largely outperforms state-of-the-artsolutions for network resource orchestration.At the time of writing DeepCog and AZTEC are, to the best of our knowledge, theonly works where a deep learning architecture is explicitly tailored to the problem ofanticipatory resource orchestration in mobile networks.
YR 2020
FD 2020-01
LK https://hdl.handle.net/10016/31013
UL https://hdl.handle.net/10016/31013
LA eng
NO Mención Internacional en el título de doctor
NO This work has been supported by IMDEA Networks Institute
DS e-Archivo
RD 27 jul. 2024