RT Conference Proceedings
T1 DeepCog: cognitive network management in sliced 5G Networks with deep learning
A1 Bega, Dario
A1 Gramaglia, Marco
A1 Fiore, Marco
A1 Banchs Roca, Albert
A1 Costa-Pérez, Xavier
AB Network slicing is a new paradigm for future 5Gnetworks where the network infrastructure is divided into slicesdevoted to different services and customized to their needs.With this paradigm, it is essential to allocate to each slice theneeded resources, which requires the ability to forecast theirrespective demands. To this end, we present DeepCog, a noveldata analytics tool for the cognitive management of resourcesin 5G systems. DeepCog forecasts the capacity needed to accommodatefuture traffic demands within individual networkslices while accounting for the operator’s desired balance betweenresource overprovisioning (i.e., allocating resources exceedingthe demand) and service request violations (i.e., allocating lessresources than required). To achieve its objective, DeepCog hingeson a deep learning architecture that is explicitly designed forcapacity forecasting. Comparative evaluations with real-worldmeasurement data prove that DeepCog’s tight integration ofmachine learning into resource orchestration allows for substantial(50% or above) reduction of operating expenses withrespect to resource allocation solutions based on state-of-theartmobile traffic predictors. Moreover, we leverage DeepCogto carry out an extensive first analysis of the trade-off betweencapacity overdimensioning and unserviced demands in adaptive,sliced networks and in presence of real-world traffic.
PB IEEE
YR 2019
FD 2019-06-17
LK https://hdl.handle.net/10016/28331
UL https://hdl.handle.net/10016/28331
LA eng
NO Proceeding of: 2019 IEEE International Conference on Computer Communications (IEEE INFOCOM 2019), Paris (France), 29 April - 2 May, 2019.
NO The work of University Carlos III of Madrid was supportedby the H2020 5G-MoNArch project (Grant Agreement No.761445) and the work of NEC Laboratories Europe by the 5GTransformerproject (Grant Agreement No. 761536). The workof CNR-IEIIT was partially supported by the ANR CANCANproject (ANR-18-CE25-0011).
DS e-Archivo
RD 7 jul. 2024