The placement of Virtual Network Functions (VNFs) plays a critical role in the provisioning of services across nextgeneration networks. However, the constantly changing conditions in network topologies, traffic demands and resource availability pose significant challenges to achieving efficient and reliable placement strategies. While Deep Reinforcement Learning (DRL) has been widely explored to address this problem, these approaches have limited generalization capabilities and require retraining when faced with unfamiliar scenarios, such as network topology changes, failure events or evolving service requests (e.g., VNF Forwarding Graphs or VNF-FGs). To address these limitations, we propose GPG-VNFE, a foundation Graph Neural Network (GNN) architecture that integrates graph contrastive learning with generative models as Learnable Priors (GraphCL-LP) within a Transformer-based DRL framework. GPG-VNFE first pretrains two domain-specific GNN encoders using GraphCL-LP, one for substrate network topologies and another for VNF-FG service graphs. Through extensive simulation on a medium-size network topology, we show that GPG-VNFE improves the acceptance ratio of incoming VNF requests by up to 57.45% compared with three baseline state-of-the-art placement approaches. We also show that our proposal lowers the power consumption at the physical layer by up to 11%, outperforming two of the baseline approaches.

The advent of Network Function Virtualization (NFV) and virtualized Content Delivery Network (vCDN) has revolutionized the deployment of resources at the edge of the network, offering a more efficient alternative to traditional CDN architectures. However, this approach introduces the challenge of resource limitations at the edge, making effective resource allocation a critical issue. This paper tackles the problem of placement of virtual network functions (VNF) by proposing a planning strategy to assign end-users access points to edge servers where vCDN functions are deployed, ensuring compliance with Service Level Agreement (SLA) while minimizing the energy consumption. We show that the problem is NP-hard and then propose a Mixed Integer Linear Program (MILP) to formulate our problem, making use of a non-linear energy model from the literature to estimate the energy footprint. We evaluate the proposal leveraging real traffic demand data from a nationwide mobile operator to model realistic network conditions. Additionally, we investigate the impact of varying the number of edge servers on the overall energy footprint. Our results demonstrate the effectiveness of the proposed optimization strategy in reducing energy consumption while maintaining the required quality of service compared to a baseline approach.

As augmented and virtual reality evolve, achieving seamless synchronization between physical and digital realms remains a critical challenge, especially for real-time applications where delays affect the user experience. This paper presents MetaLore, a Deep Reinforcement Learning (DRL) based framework for joint communication and computational resource allocation in Metaverse or digital twin environments. MetaLore dynamically shares the communication bandwidth and computational resources among sensors and mobile devices to optimize synchronization, while offering high throughput performance. Special treatment is given in satisfying end-to-end delay guarantees. A key contribution is the introduction of two novel Age of Information (AoI) metrics: Age of Request Information (AoRI) and Age of Sensor Information (AoSI) — integrated into the reward function to enhance synchronization quality. An open source simulator has been extended to incorporate and evaluate the approach. The DRL solution is shown to achieve the performance of full-enumeration brute-force solutions by making use of a small, task-oriented observation space of two queue lengths at the network side. This allows the DRL approach the flexibility to effectively and autonomously adapt to dynamic traffic conditions.

This paper introduces the design of a data pipeline system (DPS) integrated with artificial intelligence (AIF) functions to support continuous AI learning and operations for network automation in 5G/6G systems. We design the DPS as a chain of functions, namely ingress and egress Network Data Broker Function (iNDBF and eNDBF) and Network Data Preprocessing Function (NDPPF), to support in-network learning operations. To take into account the distributed nature of the network architecture of 5G systems and beyond, we conceive the DPS to be integrated seamlessly with a distributed learning frameworks such as the federated learning (FL). We performed a realistic evaluation, employing a real dataset from a national mobile operator to simulate the network architecture. Additionally, a FL framework for anomaly detection is integrated with the DPS to assess the effectiveness of our proposal. Evaluation results show that delays in end-to-end data transmission and preprocessing to the AIF locations can cause distributed learning AIFs to work with stale data. The results also highlight how the DPS can counterbalance these delays leading to desynchronisation of the distributed learning process, bringing to AIFs with higher accuracy.

Blockchains are used to store and transmit digital and secure information. This is made possible by creating a chain of chronologically and cryptographically linked data blocks. An oracle is often used to feed the blockchain to ensure the data inserted are reliable. The nature of data sent to the oracle can vary a lot from one application to another. In the scope of the Internet of Things, data provided by sensors may be corrupted when the sensor is damaged or has been corrupted. In this case, it is important for the oracle to include an efficient tool able to detect these outliers. The Grubbs test is such an efficient tool. This article aims at presenting the impact of the use of the Grubbs test on the performance of the oracle. It shows that not only it allows to significantly increase the quality of the data inserted in the blockchain by identifying outliers, it also keeps the algorithmic complexity of the oracle as low as possible.

Federated learning is experiencing great growth thanks to its many advantages, particularly in terms of security and cost reduction. With the progress of 5G/6G networks, this approach is increasingly explored for connected vehicle networks in which vehicles collaboratively build learning models under the orchestration of network base stations. However, existing works neglect the very volatile context of both radio network and vehicle mobility in the design of federated learning solution. Indeed, vehicles are not fixed entities and their residence time, i.e. connection duration to a base station, is a function of their mobility and the load of the base station. In this paper we present a federated learning approach adapted for vehicular networks. This approach is based on two models that we define: a model for predicting vehicle residence times in base stations based on their speeds and positions; a model for selecting vehicles that participate in federated learning which takes into account predicted residence times. Test results based on real data and simulation data show that our approach provides better learning performance in terms of convergence speed and precision.

Network slicing allows leveraging virtualization techniques for the creation of multiple, logically-isolated network instances over a shared infrastructure. In general, it is composed of a set of unique network functions with a specific physical capacity request. In order to improve network and service robustness, ETSI has introduced the concept of Network Function Set where network functions are replicated and deployed in different physical nodes. In this paper, we integrate this concept of the network function set to implement load-balancing and efficient resource distribution in 5G networks by leveraging on an existing network slice design formulation without the network function set. This helps relieve the burden on the nodes and links and prevent QoS degradation, in particular during failures. We describe how the baseline approach is impacted for the placement of network functions. We then show that our approach improves load balancing and latency with respect to the baseline solution.

Edge intelligence combined with federated learning is considered as a way to distributed learning and inference tasks in a scalable way, by analyzing data close to where it is generated, unlike traditional cloud computing where data is offloaded to remote servers. In this paper, we address the placement of Artificial Intelligence Functions (AIF) making use of federated learning and hardware acceleration. We model the behavior of federated learning and related inference point to guide the placement decision, taking into consideration the specific constraint and the empirical behavior of a virtualized infrastructure anomaly detection use-case. Besides hardware acceleration, we consider the specific training time trend when distributing training over a network, by using empirical piece-wise linear distributions. We model the placement problem as a MILP and we propose a variant of the problem. Simulation results show the impact that hardware acceleration can have in the decision of the number of AIF to enable, while dividing by a relevant factor the distributed training time. We also show how our approach exacerbates the importance of monitoring an end-to-end learning system delay budget composed of link propagation delay and distributed training time in the location of AIFs.

Multi-access Edge Computing (MEC) technology has emerged to overcome traditional cloud computing limitations, challenged by the new 5G services with heavy and heterogeneous requirements on both latency and bandwidth. In this work, we tackle the problem of clustering access points in MEC environments, introducing a set of clustering models to be deployed at the pre-provisioning phase. We go through extensive simulations on real-world traffic demands to evaluate the performance of the proposed solutions. In addition, we show how MEC hosts capacity violation can be decreased when integrating access points clustering into the orchestration model, by investigating on solution accuracy when applied on held-out users traffic demands. The obtained results show that our approach outperforms two state-of-the-art algorithms, reducing both memory usage and execution time, by 46% and 50%, respectively, in comparison to a baseline algorithm. It surpasses the two methods in gaining control over MEC hosts capacity usage for different maximum achieved occupancy levels on MEC hosts.