Traditional telecommunication network functions often require configuring thousands of parameters, resulting in large configuration files. The built-in capabilities of Kubernetes struggle to manage the vast number of parameters in a single configuration file, leading to workarounds that are not native to Kubernetes. This paper proposes an automated, Kubernetes-native configuration deployment procedure based on GitOps principles and the Kubernetes API aggregation layer. This approach can solve the problems originating from the existence of large configuration files. Furthermore, we show numerical results about the deployment times to illustrate the engineering cost of the new approach.
A major challenge for telecommunications operators is to dimension and plan network capacity for all types of content delivered over the network, much of this being so-called “over-the-top” (OTT) traffic from third party content providers and aggregators. While Content Delivery Networks (CDN) are a very useful means of realising efficient delivery of large volumes of traffic, they can also be limited in flexibility and cost-efficiency due to the hardware-oriented nature of the architecture. Virtual CDNs (vCDNs) enable a more flexible and optimal delivery model whereby telco compute infrastructure offers a hosting environment for softwarised CDN caches. Such an approach also lends itself to increased levels of distribution and de-centralisation; edge compute to host vCDNs can provide both service quality advantages on the one hand, and network capacity planning benefits, on the other. This latter advantage arises from the opportunity for “local turn-around” of traffic due to cache hits on popular cached content. Only real-world trials of vCDN can determine these offload benefits in quantitative terms. This paper explains a major trial with two CDN partners (one delivering video content, the other delivering game content) performed in summer 2024 for a “metro” area; significant cache efficiency levels ranging from 50% to above 90% were observed confirming that significant traffic offload and thus capacity planning benefits - can be realised.
Modern enterprises depend on uninterrupted WideArea Network (WAN) connectivity between geographically dispersed sites and cloud services. Software-Defined Wide Area Networks (SD-WANs) have emerged as a compelling solution, offering greater flexibility and cost-efficiency compared to conventional WAN technologies. Ensuring robust availability-defined as the network's readiness to provide connectivity-is crucial for a successful SD-WAN implementation. This work presents a comprehensive, model-based evaluation framework designed to quantify service-oriented availability between pairs of SDWAN endpoints. Our hierarchical modeling approach effectively addresses the complexity and accurately reflects the dependencies among various SD-WAN components. Furthermore, we investigate the impact of a centralized control plane and its associated non-idealities on availability, providing insights into communication delays, sampling frequency of network characteristics, and algorithm execution times through simulation. In-depth experiments demonstrate the evaluation capabilities of our framework, including sensitivity analysis to identify critical components, the effects of various non-idealities, and how different control plane policies can mitigate them. These results offer valuable guidance for optimizing SD-WAN design before real-world implementation.
As the demand for mobile data continues to grow, the energy consumption of mobile networks becomes a major concern. Specifically, base stations account for over 76% of energy usage in mobile networks. We consider a two-tier cellular system, one offering basic coverage while the other offers extra capacity. To save energy, existing network features can opportunistically shut down capacity layer cells when physical resources are lightly utilized. Nevertheless, this is performed without guaranteeing coverage cells can maintain the sought user-centric service quality. To address this challenge, we propose a machine learning (ML)-aided search approach that dynamically designs energy-saving configurations for each cell in each hour while being constrained with a pre-defined quality of service (QoS) measure. This ML model was trained using data collected from 10,283 cells of a live network. We introduce two different approaches to provide these settings: Adaptive QoS Threshold Optimization Algorithm (AQTOA) and an Exhaustive Search (ES) baseline. AQTOA is a low complexity ML-aided search algorithm designed to determine the optimal shutdown threshold for capacity cells while ensuring that QoS requirements are met. Through extensive live network experimentation, the AQTOA results indicate a 1.8% improvement in energy savings compared to the earlier static settings models while maintaining a more strict QoS level than the one addressed in the previous work.
Virtual reality (VR) cloud gaming faces significant challenges in terms of latency and bandwidth consumption, particularly in softwarized 5 G and beyond (5G&B) networks. Ensuring an immersive experience on low-end devices, such as smartphones and cardboard-like headsets, requires innovative solutions to mitigate delay effects and optimize resource usage. This research focuses on enhancing VR cloud gaming performance by leveraging Digital Twins (DTs) and hierarchical compression techniques. The proposed system adopts a client-server architecture, where computationally intensive tasks are offloaded to the server, minimizing processing requirements on the client side. A key contribution is a novel delay estimation and management framework that employs DTs to synchronize video streaming and user input, reducing the impact of latency. Additionally, viewport-based hierarchical compression is applied to optimize bandwidth consumption while maintaining high perceived quality. The proposed architecture is evaluated through extensive experiments analyzing the effects of input delay, delay fluctuations, and compression methods on user performance and perceived quality. Results demonstrate that the system effectively mitigates adverse network conditions, ensuring a seamless and immersive VR gaming experience even in suboptimal environments.
With the advent of next-generation networks, new applications across multiple domains are gaining traction. This shift demands a redefined network paradigm, where ultrareliable, low-latency communication is key. In this work, we explore and extend the concept of in-network programmability in new directions. Unlike conventional approaches, we leverage P4 data plane programmability to implement an in-network collision avoidance algorithm in a UAV scenario. We evaluate our hardware-based implementation under different conditions, including latency and velocity, demonstrating that it efficiently detects and prevents collisions. Our results show the impact of end-to-end latency and highlight how in-network processing can be a valuable ally for time-sensitive tasks, paving the way for future advancements in hardware-based in-network applications.
Network management systems for beyond 5G (B5G) and 6G networks today require efficient approaches for handling increased heterogeneity, network-function dis-aggregation, performance requirements, and optimizing networks to support highly diverse use cases. While the Open-Radio Access Network (O-RAN) Service Management and Orchestration (SMO) frameworks efficiently manage RAN and cloud infrastructure, the fragmentation of management platforms across RAN, Core Network (CN), and Transport Network (TN) introduces operational inefficiencies, particularly in Non-Public Networks (NPNs) deployments. This paper proposes a novel extension to the O-RAN SMO architecture, integrating CN and TN management functionalities into a unified control framework. By exploiting AI/ML-driven x/rApps and a converged data analytics pipeline, the proposed architecture enhances SMO's fault management, resource optimization, and service continuity capabilities. Our implementation validates the feasibility of the proposed SMO extension, demonstrating subscriber-specific QoS assurance through ORAN-based mobility management mechanisms. The proposed approach successfully shows how RAN-/Core-converged SMOs enable significant enhancements to O-RAN's x/rApps, allowing for subscriber-specific as well as application-specific differentiated QoS assurance.
The advent of fifth-generation mobile networks (5G) has brought transformative improvements in data transfer speeds, latency reduction, and device connectivity, enabled by the softwarization, virtualization, and disaggregation of network functions. Open-source projects have been instrumental in facilitating software-driven 5G deployments on standard hardware, accelerating the development of private 5G networks, test environments, and advancing research and innovation in the domain. Despite these advancements, the current maturity of open-source solutions remains inadequate for high-performance, large-scale 5G deployments. Existing 5G performance testing approaches primarily adopt a black-box perspective, which fails to capture the intricate interactions among core 5G components, thereby hindering the identification of performance bottlenecks. Additionally, testing is often limited to small-scale environments, and publicly available datasets from real-world deployments are scarce. To overcome these challenges, this paper introduces an observability-driven methodology for testing $5G$ core network performance, offering deeper insights into the internal interactions of network components. We contribute to research and innovation in 5G and beyond field by (i) enabling the simulation of varying user arrival rates, (ii) releasing a public dataset capturing data of 5G core components interactions, logs, and performance metrics, and (iii) providing a comparative analysis of two popular open-source 5G core implementations based on our observability-oriented methodology.
The advent of 6G networks is driving a paradigm shift toward AI-native infrastructures, embedding artificial intelligence deeply across multiple operational layers. This evolution necessitates transparency, trustworthiness, and explainability to foster reliable and accountable network decisions. Explainable AI (XAI) techniques have emerged as crucial tools to ensure that AI-driven network operations are interpretable and aligned with human expectations, thus addressing inherent reliability issues like model degradation and risks associated with large language models (LLMs), including hallucinations. Complementing these advancements, Intent-Driven Management (IDM), standardized by 3GPP, enables translation of high-level user intents into actionable, automated network decisions, enhancing flexibility, efficiency, and resilience. This paper presents a comprehensive end-to-end (E2E) Explainability Orchestration framework integrated across resource, service, and business levels, guided by an intent-based approach. Aligned with ITU-R and 3GPP SA1 visions, the proposed AI-native orchestration framework leverages real-time analytics, intent-driven automation, and integrated explainability to ensure network transparency, regulatory compliance, and trustworthiness in next-generation telecommunication ecosystems.
The evolution of cloud computing towards a cloud continuum, including cloud, edge, and far-edge resources, is revolutionizing the deployment, management, and orchestration of Network Services (NSs) and applications. Traditional, centralized orchestration approaches are increasingly inadequate for handling the complexity, scale, and dynamic nature of this continuum. In this paper, we present a data-driven approach for AI-powered service orchestration based on the European 6G-CLOUD project. Specifically, we introduce the Decentralized Service Orchestrator (DSO) framework, an AI-powered, decentralized orchestration model that leverages the capabilities of the Artificial Intelligence and Machine Learning Framework (AI/MLF) to enable intelligent, autonomous, and scalable service lifecycle management across heterogeneous environments. Key contributions include the detailed architecture of the DSO, its workflows, and its integration with the Cloud Continuum and with an AI/MLF that manage the AI lifecycle, enabling models provision to the different components. By enabling decentralized AI-driven decision-making, this framework enhances service reliability, scalability, operational efficiency, and innovation acceleration, paving the way for next-generation cloud continuum orchestration.
Remote Direct Memory Access (RDMA) is a key technology in modern data centers, enabling low-latency and high-throughput communication. However, evaluating RDMA performance and validating network designs often requires costly hardware setups or simulation tools with limited performance and realism. In this work, we present P4DMA, a system that leverages programmable switch ASICs to generate realistic high-performance RDMA traffic. By implementing RDMA traffic patterns using the P4 language, P4DMA enables researchers and practitioners to generate RDMA workloads at line rate, without relying on traditional RDMA NICs (RNICS). With Tofino's traffic generation capacity of up to Tbps, P4DMA offers a novel approach to stress and evaluate RDMA-capable infrastructures, and accelerate the prototyping of new RDMA-based applications.
This paper proposes an intent-based management framework for orchestrating In-Network Computing Functions (ICFs) in Cooperative Intelligent Transportation Systems (C-ITS). Software-Defined Vehicles (SDVs) are emerging as key enablers of autonomous driving in C-ITS. SDVs leverage cloud-native platforms (e.g., Kubernetes) with their own in-vehicle networks and Vehicle-to-Everything (V2X) communication device. The proposed framework enables high-level intent-driven configuration of networking, security, and application services across SDVs, infrastructure nodes, and Mobile Network Operators (MNOs). A prototype implementation of the ICF framework demonstrates the feasibility of its architecture. This paper discusses several design challenges to guide future research in scalable, intentbased orchestration for SDV ecosystems.
While the O-RAN Alliance is gaining momentum in democratizing RAN innovation through multi-vendor interoperability, there remains a gap in empirical performance evaluation of O-RAN applications deployed with future-proof security measures on carrier-grade components. Furthermore, how implementing an advanced security measure impacts the overall O-RAN performance is largely understudied in true cloud-native deployments. This paper presents a comprehensive integration of an O-RAN-specific testbed that utilizes a quantum-safe security technology, an industry-leading O-RAN test emulator, and a carrier-grade Near-Real-Time (near-RT) RIC platform hosting our custom AI-driven Traffic Steering (TS) xApp. Our threefold contribution includes developing an innovative TS xApp for efficient RAN load-balancing on the fly, empirically measuring RAN key performance indicators (KPIs), and assessing the security solution's impact on overall system performance. Results demonstrate that our TS xApp enhances the downlink throughput by 3.88% and improves the median user QoS scores by 2.52%. In particular, the quantum-safe security measure implemented to secure the O-RAN E2 interface in the multi-vendor deployment, adds only 0.818 milliseconds of average latency which remains well below one second, the maximum acceptable threshold for near-RT RIC control loop response time.
Modern computer networks, such as 5G/6G networks, require high-performance, low-latency, and secure packet processing while ensuring data confidentiality in cloud environments. Trusted Execution Environments (TEEs) address these security requirements and provide encrypted memory areas that protect sensitive data from untrusted cloud providers. This paper presents a performance analysis of TEE technologies, specifically Intel SGX and AMD SEV-SNP, in the context of software-based user-space packet processing with DPDK and the P4 language. We evaluate two architectural approaches: (1) integrating TEEs as external processing modules implemented with SGX and (2) executing the entire P4 pipeline inside a TEE using AMDSEV. Our analysis examines computational and I/O overhead across different CPU architectures. The results show the tradeoffs between TEE designs, implementations, and performance, demonstrating that AMD SEV-SNP offers better scalability with lower performance penalties compared to Intel SGX.
With the high demands of 6 G networking services in terms of Quality of Service (QoS), managing these networks requires intelligent next-generation Operations Support Systems (OSSs). According to standardization bodies such as ETSI and 3GPP, OSS must support end-to-end, cross-domain management across all 6 G domains. They are making significant efforts to standardize Application Programming Interfaces (APIs) to enable Intent-Based Networking (IBN), which simplifies network management by allowing users to express their intentions in a declarative manner. However, these systems remain complex for users with limited domain knowledge who need to interact with these standardized APIs. Moreover, adding new functionalities to OSS often requires users to learn new API endpoints and structures, which can be time-consuming. To address these challenges, we propose enabling natural language interaction with OSS by leveraging Large Language Models (LLMs). Our approach offers two key advantages: simplifying user interaction with the system using natural language, and enabling the system to autonomously adapt to new API features. Since fulfilling a user's intent may involve multiple low-level API calls, our solution is designed to plan and execute them in a coordinated manner. We employ multi-agent LLMs with a hierarchical planning mechanism, creating a chatbot-like system that processes natural language inputs effectively. Real-world experiments conducted at EURECOM's OSS demonstrated that the proposed approach can efficiently manage all 6 G domains using natural language.
Holographic-type communication demands multi-Gbps throughput and ultra-low latency, posing significant challenges to current 5G networks. This demo work presents a programmable testbed built on the P7 network emulator, which overcomes the throughput constraints of conventional platforms. Our demonstration deploys three edge computing slices—one at a far edge and two at near edges—to support volumetric media streaming directly to VR head-mounted display. By facilitating precise adjustment of network metrics such as latency and packet loss across the P7-enabled slices, the testbed allows users to visually observe the resulting effects on holographic content in VR. Although automated measurement of Quality of Experience (QoE) parameters are not incorporated, the testbed provides a valuable experimental environment where researchers can subjectively evaluate the relationship between network conditions and perceived quality, thereby gaining insights into optimizing network performance for enhanced Quality of Service (QoS).
The 5G/6G era has introduced a wide variety of services, including enhanced Mobile Broadband (eMBB), UltraReliable Low-Latency Communications (URLLC), and massive Machine-Type Communications (mMTC). Each service presents unique, highly diversified, and often conflicting requirements, driving the need for more flexible and intelligent solutions. In this context, Network Slicing (NS) has emerged as a prominent technology that allows multiple virtual networks to operate over a shared physical infrastructure, thereby accommodating these diverse service demands. Supported by technologies such as Software-Defined Networking (SDN) and Network Function Virtualization (NFV), network slicing requires the efficient placement of slices to optimize resource utilization and ensure Quality of Service (QoS). We propose a native artificial intelligence (AI) architecture for end-to-end (E2E) slicing that leverages Transformer-based Deep Reinforcement Learning (DRL) to enable zero-touch, automated slice placement in future networks, such as 5 G -and-beyond systems. Our system embeds AI directly into the network fabric, supporting native AI for real-time data processing and decision-making. Results show that integrating the Transformer model with DRL effectively addresses complex optimization challenges in network slicing, outperforming other state-of-the-art learning algorithms by better balancing slice acceptance ratio and energy efficiency. This supports the sustainable management of future networks, aligns with the vision of the Next Generation Mobile Networks (NGMN) Alliance, and illustrates the evolving role of AI in next-generation communication systems.
WiFi Access Points (APs) process and switch packets between the WiFi and Ethernet interfaces. These operations are managed by the Linux kernel's data plane; however, the kernelbased packet switching is inherently complex and challenging to monitor and modify. To address the growing demands for WiFi applications and services, and in particular to foster innovation in advancements of the data plane for WiFi APs, this work introduces AirXDP, a user-space packet switching architecture that leverages Linux's eXpress Data Path (XDP) framework. In this paper, we first present the AirXDP system architecture and the details of packet processing engine in the user-space. Next, we identify and address key limitations, such as the overhead caused by excessive polling of ingress queues as well as the increased RTTs resulting from standing queues. Through empirical and analytical evaluations, we demonstrate that techniques such as optimizing polling intervals, configuring core affinities, efficiently managing queues, and using XDP's native attach mode even on one interface enable AirXDP to achieve higher processing efficiency compared to packet switching performed by the Linux kernel. Overall, AirXDP provides an efficient packet switching framework for facilitating the enhancement of WiFi AP's data plane.
As the requirement for flexibility in networks grows, new mechanisms are needed to enable extreme adaptability to diverse environments and the ability to react to unknown situations. The rise of AI provides a significant opportunity to enhance network efficiency, scalability, and resilience beyond what is possible with traditional policy-based approaches. However, current network architectures rely on predefined rules and event-triggered responses, insufficient to dynamically address unpredictable and diverse conditions. This paper introduces a new perspective on AI-driven autonomous network agents, capable of continuous perception, proactive decision-making, and adaptive optimization across multiple network layers. These agents automate deployment-phase tasks, optimize runtime operations, and enhance cross-layer coordination, addressing critical challenges like mobility management, resource scheduling, and dynamic service adaptation. Unlike traditional static policies, these agents learn from operational feedback, enabling networks to adjust and refine their behavior over time. With their self-learning and intent-driven automation, AI-driven agents integrated into future network architectures, including beyond 5G and 6G systems, can enable greater autonomy, adaptive optimization, and improved robustness, effectively responding to evolving communication requirements.
Interpreting complex 3GPP telecommunications standards for question and answering (QA) poses a challenge for general-purpose LLMs due to their specialized terminology and high computational demands, limiting their use in resourceconstrained environments. This work explores an efficient, opensource approach using the TeleQnA dataset of 10,000 telecom questions and the TSpec-LLM repository of processed 3GPP documents. We enhance a lightweight Llama 3.2 (3B parameters) model, quantized from 16-bit precision to 4 bits, through finetuning and RAG to improve accuracy without heavy resource reliance. Unlike prior resource-intensive or proprietary solutions, our method reduces memory demands, enabling deployment on modest hardware like edge devices or softwarized networks. Shared via GitHub repositories [1], this approach advances costeffective, reproducible AI for telecommunications QA, supporting contexts where budgets, computation, or public internet access are limited.
