ABOUT
Dr. Banchs received his MSc and PhD degrees from UPC BarcelonaTech in 1997 and 2002. He was a visitor researcher at ICSI, Berkeley, CA, in 1997, worked for Telefonica I+D, Spain, in 1998, and for NEC Network Laboratories, Germany, from 1998 to 2003. Since 2003 he is with Carlos III University of Madrid and since 2009 he has a double affiliation as Deputy Director of the IMDEA Networks institute. He was Academic Guest at ETHZ in 2012, Visiting Professor at EPFL in 2013, 2015 and 2018, and Fulbright scholar at University of Texas at Austin in 2019. He authors more than 100 publications, is editor for IEEE/ACM Transactions on Networking and has been editor and TPC member for many journals and conferences. He has received a number of awards, including several paper and project awards, the national prize to the best PhD thesis on broadband networks and the 2013 and 2017 Excellence Awards to Young Research Staff at Carlos III University of Madrid. He was an IEEE Distinguished Lecturer in 2016-2017.
Tutorial: Data-analytics based orchestration
Abstract: The telecom industry is moving from a “horizontal” service delivery model, where services are defined independently from their consumers, towards a “vertical” delivery model, where the provided services are tailored to specific industry sectors and verticals. In order to enable this transition, a 5G architecture is needed with capabilities to support the use cases of the different vertical industries. A key feature of this architecture is the implementation of network slicing over a single infrastructure to provision highly heterogeneous vertical services, as well as a network slicing management system capable of handling simultaneous slices. On top of the network slicing technology, functionality needs to be devised to deploy the slices required by the different vertical players and provide them with a suitable interface to manage the slice. This includes the MANO and control functionality as well as artificial intelligence-based data analytics.
In order to orchestrate a network slice to meet the requirements of vertical customers, we need to select the most appropriate location for the different virtual network functions (VNFs) instantiated for a network slice, including the ones corresponding to application layer functions that can be located within the network relying on the multi-access edge computing (MEC) capabilities of the infrastructure. This involves an orchestration functionality that takes as input from the underlying infrastructure the resource availability in the different nodes of the network, accounting both for computing and communications resources. Based on the output, the Orchestrator instantiates the network slice as follows: (i) it issues the corresponding requests to the Software-Defined Networking (SDN) controller to instantiate connections between the different network nodes, (ii) it requests the Virtualized Infrastructure Manager (VIM) to reserve the virtual resources at the different network nodes, (iii) it commands the Virtual Network Function Manager (VNFM) to instantiate the VNFs, and (iv) it configures the VNFs and PNFs (Physical Network Functions).
For the orchestration and control functionality, Artificial Intelligence (AI) plays a very prominent role, featuring an AI-based Data Analytics framework that allows for autonomous and more efficient control, management and orchestration of the network. The framework is particularly well aligned with the concepts defined by 3GPP for the Network Data Analytics Function (NWDAF) of the 5G Core (5GC) and it incorporates the basics of the functionality defined for the Management Data Analytics Function (MDAF). The AI-based data analytics framework relies on control plane data analytics for the optimization of the 5GC NFs. It further utilizes an orchestrator and management system that relies also on analytics for making its decisions, enhancing the Communication Service Management Function (CSMF) and the Network Slice Management Function (NSMF).
In the management and orchestration planes, the data used as input by the AI-based analytics framework is provided by the NFV Infrastructure (NFVI) and the MANO system. The NFVI information provides the knowledge on the computational resources’ capabilities (such as the type of CPU and memory, accelerators, etc.) along with their availability (i.e., the status and utilization level). Based on this information and running AI-based algorithms, the framework influences and optimizes the placement decisions made by the VIM, while ensuring that the resulting resource allocation satisfies the respective slice SLAs. AI-based data analytics is also employed to optimize resource provisioning and admission control of network slices.
Keynote: Resource Allocation for network slicing
Abstract: There is consensus among the relevant industry and standardization communities that a key element in 5G mobile networks is network slicing. The idea is to allow the mobile infrastructure to be “sliced” into logical networks, where each slice is a collection of resources and functions that includes software modules running at different locations as well as the nodes’ computational and communication resources. By providing specially tailored instances of network slices, this allows for a strong specialization of the offered services on the same shared infrastructure. Network slicing has profound implications on resource management, as it entails an inherent trade-off between (i) the need for fully dedicated resources to support service customization, and (ii) the dynamic resource sharing among services to increase resource efficiency and cost-effectiveness of the system. In this talk, we investigate this trade-off, analyzing the efficiency gap introduced by non-reconfigurable allocation strategies of different kinds of resources, from radio access to the core of the network, and showing the advantages of their dynamic orchestration at different timescales. We further present mechanisms that realize the potential of dynamic orchestration by allocating to each slice the needed resources at every point in time, which requires the ability to forecast the demands of the different network slices over time. Machine learning appears as a natural tool to this end. We present a novel data analytics tool for the cognitive management of resources in 5G systems that hinges on a deep learning architecture, accounting for the operator’s desired balance between resource overprovisioning (i.e., allocating resources exceeding the demand) and service request violations (i.e., allocating less resources than required).