The high energy footprint of 5G base stations, particularly the radio units (RUs), poses a significant environmental and economic challenge. We introduce Kairos, a novel approach to maximize the energy-saving potential of O-RAN's Advanced Sleep Modes (ASMs). Unlike state-of-the-art solutions, which often rely on complex ASM selection algorithms unsuitable for time-constrained base stations and fail to guarantee stringent QoS demands, Kairos offers a simple yet effective joint ASM selection and radio scheduling policy capable of real-time operation. This policy is then optimized using a data-driven algorithm within an xApp, which enables several key innovations: ($i$) a dimensionality-invariant encoder to handle variable input sizes (e.g., time-varying network slices), (ii) distributional critics to accurately model QoS metrics and ensure constraint satisfaction, and (iii) a single-actor-multiple-critic architecture to effectively manage multiple constraints. Through experimental analysis on a commercial RU and trace-driven simulations, we demonstrate Kairos's potential to achieve energy reductions ranging between 15% and 72% while meeting QoS requirements, offering a practical solution for cost- and energy-efficient 5G networks.

The virtualization of Radio Access Networks (vRAN) is well on its way to become a reality, driven by its advantages such as flexibility and cost-effectiveness. However, virtualization comes at a high price — virtual Base Stations (vBSs) sharing the same computing platform incur a significant computing overhead due to in extremis consumption of shared cache memory resources. Consequently, vRAN suffers from increased energy consumption, which fuels the already high operational costs in 5G networks. This paper investigates cache memory allocation mechanisms’ effectiveness in reducing total energy consumption. Using an experimental vRAN platform, we profile the energy consumption and CPU utilization of vBS as a function of the network state (e.g., traffic demand, modulation scheme). Then, we address the high dimensionality of the problem by decomposing it per vBS, which is possible thanks to the Last-Level Cache (LLC) isolation implemented in our system. Based on this, we train a vBS digital twin, which allows us to train offline a classifier, avoiding the performance degradation of the system during training. Our results show that our approach performs very closely to an offline optimal oracle, outperforming standard approaches used in today’s deployments.

Radio Access Networks virtualization (vRAN) is on its way becoming a reality driven by the new requirements in mobile networks, such as scalability and cost reduction. Unfortunately, there is no free lunch but a high price to be paid in terms of computing overhead introduced by noisy neighbors problem when multiple virtualized base station instances share computing platforms. In this paper, first, we thoroughly dissect the multiple sources of computing overhead in a vRAN, quantifying their different contributions to the overall performance degradation. Second, we design an AI-driven Radio Intelligent Controller (AIRIC) to orchestrate vRAN computing resources. AIRIC relies upon a hybrid neural network architecture combining a relation network (RN) and a deep Q-Network (DQN) such that: (i) the demand of concurrent virtual base stations is satisfied considering the overhead posed by the noisy neighbors problem while the operating costs of the vRAN infrastructure is minimized; and (ii) dynamically changing contexts in terms of network demand, signal-to-noise ratio (SNR) and the number of base station instances are efficiently supported. Our results show that AIRIC performs very closely to an offline optimal oracle, attaining up to 30% resource savings, and substantially outperforms existing benchmarks in service guarantees.

Network slicing is well-recognized as a core 5G technology to enable heterogeneous vertical services sharing the same infrastructure. In this context, the H2020 5Growth project extends baseline 5G management and orchestration platforms to manage the life-cycle of real end-to-end, reliable, and secure network slices with performance guarantees. In this paper, we present 5Growth's approaches to ($i$) attain isolation across network slices, (ii) provide secure interfaces towards third parties, and (iii) exploit AI/ML to achieve reliability through automated anomaly detection. In our quest towards validating full-fledged 5G pilots, we demonstrate our slicing mechanisms in PoCs that include interacting with ICT-17 infrastructure.

The novel concept of network slicing is envisioned to allow service providers to open their infrastructure to vertical industries traditionally alien to mobile networks, such as automotive, health or factories. In this way multiple vertical services can be delivered over the same physical facilities by means of advanced network virtualization techniques. However, the vertical service requirements heterogeneity (e.g., high throughput, low latency, high reliability) calls for novel orchestration solutions able to manage end-to-end network slice resources across different domains while satisfying stringent service level agreements. In this demonstration we will show a novel orchestration solution able to handle one of the most stringent requirements: end-to-end latency. Our testbed-evolution of the work presented in [1]-implements all the resource brokerage schemes and allocation operations necessary to complete the life-cycle management of network slices. In addition, the novel overbooking concept is applied to pursue the overall revenue maximization when admitting network slices. Finally, an advanced network slicing monitoring system will be provided as a user-friendly dashboard allowing users to interact with the proposed solution.

5G networks will pose complex network management challenges due to the variety of vertical services they will need to serve and the diversity and heterogeneity of underlying infrastructure. The service orchestration functionality is fundamental to enable fulfilling the requirements of the different verticals while efficiently sharing the infrastructure resources. This paper details the 5G-TRANSFORMER service orchestrator implementation and operation. It also evaluates and profiles service creation time showing how the automation offered by the platform allows reducing it from hours to minutes. It also shows that the most time-consuming steps correspond to the deployment of the virtual network functions and post-deployment configuration, which consume one order of magnitude more time than the rest of steps (e.g., network creation, port creation).

The goal of 5G-Crosshaul is to integrate fronthaul and backhaul operation under the same data and control planes. This paper focuses on the latter, by experimentally showing the flexibility of the 5G-Crosshaul Control Infrastructure (XCI). In this sense, various network setups featuring heterogeneous network and computing resources and high-speed mobility were deployed over the 5G-Crosshaul testbed. More specifically, three different use cases that exploit the capabilities embedded in the XCI have been experimentally evaluated. First, "hierarchical network orchestration" demonstrates how service setup times in complex multi-technology transport networks can be decreased from current manual configuration times in the order of days down to automated setups in the order of seconds by means of a resource management application that consumes the XCI services. Second, "energy management of IT and network resources" presents an energy management application that exploits the XCI to deploy network configurations that achieve energy savings ranging from 15% to 40% by dynamically reacting to datacenter and network conditions. Finally, the XCI was also exploited by an energy management application in a high-speed train mobility scenario featuring a radio over fiber network in which savings close to 80% were achieved.

Cloud Radio Access Network (C-RAN) will become a main building block for 5G. However, the stringent requirements of current fronthaul solutions hinder its large-scale deployment. In order to introduce C-RAN widely in 5G, the next generation fronthaul interface (NGFI) will be based on a cost-efficient packet-based network with higher path diversity. In addition, NGFI shall support a flexible functional split of the RAN to adapt the amount of centralization to the capabilities of the transport network. In this paper we question the ability of standard techniques to route NGFI traffic while maximizing the centralization degree-the goal of C-RAN. We propose two solutions jointly addressing both challenges: (i) a nearly-optimal backtracking scheme, and (ii) a low-complex greedy approach. We first validate the feasibility of our approach in an experimental proof-of-concept, and then evaluate both algorithms via simulations in large-scale (real and synthetic) topologies. Our results show that state-of-the-art techniques fail at maximizing the centralization degree and that the achievable C-RAN centralization highly depends on the underlying topology structure.

Network slicing allows mobile operators to offer, via proper abstractions, mobile infrastructure (radio, networking, computing) to vertical sectors traditionally alien to the telco industry (e.g., automotive, health, construction). Owning to similar business nature, in this paper we adopt yield management models successful in other sectors (e.g. airlines, hotels, etc.) and so we explore the concept of slice overbooking to maximize the revenue of mobile operators. The main contribution of this paper is threefold. First, we design a hierarchical control plane to manage the orchestration of slices end-to-end, including radio access, transport network, and distributed computing infrastructure. Second, we cast the orchestration problem as a stochastic yield management problem and propose two algorithms to solve it: an optimal Benders decomposition method and a suboptimal heuristic that expedites solutions. Third, we implement an experimental proof-of-concept and assess our approach both experimentally and via simulations with topologies from three real operators and a wide set of realistic scenarios. Our performance evaluation shows that slice overbooking can provide up to 3x revenue gains in realistic scenarios with minimal footprint on service-level agreements (SLAs).

Future 5G deployments will support a flexible split of Base Station (BS) functions, i.e., it will be possible to decide which atomic operations will be co-located on the edge and which ones will be processed on a Central Unit (CU). Thus, network owners will be able to decide how much centralization they would like to retain in different deployments. However, deciding which BS components should be offloaded to a CU becomes a challenge because routing and BS function placement choices are coupled. We present WizHaul, a software framework enabling the implementation of a centralized functional split decision- making engine for future 5G networks. The purpose of WizHaul is twofold. First, it may be used in a network planning phase to settle the optimal amount of centralization. Second, it may also be used to support network automation/adaptation scenarios where network failures or congestion in the cloud may draw the current configuration infeasible.

Upcoming 5G mobile networks are addressing ambitious KPIs not just in terms of capacity and latency, but also in terms of network control and management. In this direction, network management schemes need to evolve to provide the required flexibility, and automated and integrated management of 5G networks. This also applies to the 5G-Crosshaul transport network, which provides an integrated fronthaul and backhaul. Software defined networking and NFV are seen as key enablers for that. This article validates the flexibility, scalability, and recovery capabilities of the 5G-Crosshaul architecture in a testbed distributed geographically. More specifically, the central component of the validation is the hierarchical 5G-XCI, conceived to handle multi-domain multi-technology transport network resources. Its performance is characterized through two experimental case studies. The first one illustrates the automated provisioning of all network resources required to deploy a complete LTE virtual mobile network featuring fronthaul and backhaul configurations. This takes 10.467 s on average for the network under test. The second one exploits the flexibility of the hierarchical XCI to apply local or centralized service recovery in the event of link failure depending on the desired path optimality vs. recovery time trade-off. On average, recovery takes 0.299 s and 6.652 s, respectively. Overall, the proposed solution contributes to attaining the target set for 5G networks of reducing service setup from hours to minutes.

The novel network slicing paradigm allows service providers to open their infrastructure to new business players such as vertical industries. In this demo, we showcase the benefits of our proposed end-to-end network slicing orchestration solution that blends together i) an admission control engine able to handle heterogeneous network slice requests, ii) a resource allocation solution across multiple network domains: radio access, edge, transport and core networks and iii) a monitoring, forecasting and dynamic configuration solution that maximizes the statistical multiplexing of network slices resources. Our orchestration solution is operated through a dashboard that allows requesting network slices on-demand, monitors their performance once deployed and displays the achieved multiplexing gain through overbooking.

This demo presents an end-to-end hierarchical control of a multi-domain multi-technology integrated fronthaul and backhaul network where a parent controller coordinates the operation of a wireless and optical child controllers managing real wireless and optical data plane resources. The services offered by this hierarchical control infrastructure are accessed through the offered REST-based northbound API by the resource management application (RMA), which is able to determine the appropriate network resource allocation based on the requested traffic profiles and the position of the different mobile entities deployed in the network. Interestingly, the same REST-based API is used between the parent and child controllers inside the control infrastructure, allowing recursive and scalable deployments. The RMA manages fronthaul and backhaul traffic generated by a flexible mobile network deployment featuring user equipments, a remote radio head, a baseband unit, an eNodeB, and an EPC.

5G-Crosshaul aims at reducing network costs by designing an integrated transport (fronthaul/backhaul) network able to fulfill 5G requirements. Such transport networks will consist of heterogeneous technologies that need end-to-end orchestration. In this paper, we evaluate a hierarchical resource management framework for multi-domain wireless/optical networks. More specifically, we deploy a hierarchical 5G-Crosshaul Control Infrastructure (XCI) where child controllers deal with the specificities of each technology whilst the parent controller is in charge of offering to a resource management application (RMA) the appropriate abstraction level and an end-to-end view. To understand the end-to-end behavior related with service setup, we evaluate each network segment (wireless and optical), each plane (application and control planes), and each layer of the hierarchy inside the XCI. In particular, we evaluate the aggregated path setup time (in the order of seconds) as well as each component (wireless domain contributes with tens of ms and multi-layer optical network with hundreds of ms per layer for a total in the order of seconds). Path restoration results reveal the importance of leveraging control of child controllers when requiring fast response to unexpected data plane events, since an important part of the setup delay observed is due to the RMA-parent-child controller interaction and sequential message handling.

[ANGLES] How can we build a system for traffic classification so that different QoS levels can be identified? How can we analyse different network parameters and identify a specific QoS for every flow? The purp ose of this work is to build a system capable of classifying traffic according to flows' different QoS requirements. Next, classification results are going to b e used to improve the state-of-the-art traffic engineering techniques in SDN networks. As in SDN networks ele- ments are programmable from the controller, different QoS paths can b e implemented so that different traffic flows can see diferent QoS. This traffic classification system go es b eyond the classical state-of-the-art classification into mice/elephant flows. We want to identify different priority classes so that higher priority classes are able to access to more network resources. We seek improving the overall network p erformance. The approach for tackling this problem would be somehow equal to the approach used for building a new machine learning system. We are going to define a set of variables, measure them, and classify traffic according to the values of that variables into different classes. We will start defining a measurement layer for SDN. To classify trafic we first need to observe what prop erties is exhibiting a traffic flow. This first layer will enables us to measure different flow prop erties as well as p erforming other management tasks. We are going to define a layer so that we could gather flow information with minimal network over-head. This system will b e implemented distributely in the switches and will leverage flow tables switches to improve its functionalities. Second we are going to use the information captured to calculate various flow statistical fingerprints so that we can infer their network resources requirements. We will target only QoS signifficant flows. We are going to develop a machine learning algorithm in the scop e of semi-sup ervised learning that would learn form lab elled and unlab elled data to infer the most likely QoS. In particular, we are going to apply a machine learning algorithm known as Laplacian SVM to classify various network traffic flows into different QoS classes. Finally, we will prototyp e this algorithm and try it in a real world data set.