While traditional handovers (THOs) have served as a backbone for mobile connectivity, they increasingly suffer from failures and delays, especially in dense deployments and high-frequency bands. To address these limitations, 3GPP introduced Conditional Handovers (CHOs) that enable proactive cell reservations and user-driven execution. However, both handover (HO) types present intricate trade-offs in signaling, resource usage, and reliability. This paper presents unique, countrywide mobility management datasets from a top-tier mobile network operator (MNO) that offer fresh insights into these issues and call for adaptive and robust HO control in next-generation networks. Motivated by these findings, we propose CONTRA, a framework that, for the first time, jointly optimizes THOs and CHOs within the O-RAN architecture. We study two variants of CONTRA: one where users are a priori assigned to one of the HO types, reflecting distinct service or user-specific requirements, as well as a more dynamic formulation where the controller decides on-the-fly the HO type, based on system conditions and needs. To this end, it relies on a practical meta-learning algorithm that adapts to runtime observations and guarantees performance comparable to an oracle with perfect future information (universal no-regret). CONTRA is specifically designed for near-real-time deployment as an O-RAN xApp and aligns with the 6G goals of flexible and intelligent control. Extensive evaluations leveraging crowdsourced datasets show that CONTRA improves user throughput and reduces both THO and CHO switching costs, outperforming 3GPP-compliant and Reinforcement Learning (RL) baselines in dynamic and real-world scenarios.

We present TSGFM, a Time Series Graph Foundation Model for zero-shot network monitoring, leveraging spatiotemporal Graph Neural Networks (GNNs) to extract transferable representations across diverse multivariate time series (MTS) domains. Pretrained on heterogeneous time series datasets, TSGFM enables generalization without task-specific fine-tuning, addressing core challenges in dynamic network environments. TSGFM is benchmarked across five real-world MTS datasets and seven zero-shot forecasting scenarios, outperforming five state-of-the-art baselines in six out of seven tasks. Most notably, in zero-shot network monitoring analysis, TSGFM surpasses all competing models by at least 18%, even without any prior exposure to network monitoring data. We further compare TSGFM against leading Time Series Foundation Models (TSFMs), including TimeGPT and TimesFM. TSGFM achieves performance on par with TimeGPT, occasionally surpassing it, and consistently outperforms TimesFM, while using significantly less pretraining data and relying on a much simpler architecture. A detailed analysis of TSGFM’s learned spatial attention patterns reveals domain-specific connectivity structures. In particular, lower attention weights in network monitoring tasks suggest that dense spatial graphs may be unnecessary, opening opportunities for efficient spatial pruning without sacrificing accuracy. This challenges prevailing assumptions favoring fully connected spatiotemporal GNNs. To foster transparency and reproducibility, we release the complete implementation of TSGFM as open source, as well as the tested datasets.

Congestion Control (CC) plays a fundamental role in optimizing traffic in Datacenter Networks (DCNs). Currently, DCNs implement two main CC protocols: DCTCP and DCQCN. Both protocols are based on Explicit Congestion Notification (ECN), where switches mark packets when they detect congestion. Nowadays, network experts carefully set ECN parameters to optimize the average network performance. However, today’s DCNs experience rapid and abrupt changes that severely affect the network state (e.g., dynamic workloads, incasts), which leads to under-utilization and sub-optimal performance. In this paper we present GraphCC , a framework for in-network CC optimization. GraphCC relies on Multi-agent Reinforcement Learning (MARL) and Graph Neural Networks (GNN), and is compatible with widely deployed ECN-based CC protocols. The proposed solution deploys distributed agents on switches that communicate with their neighbors to cooperate and optimize the global ECN configuration. In our evaluation, we test GraphCC with three real-world traffic workloads, focusing on its capability to accommodate scenarios unseen during training (e.g., traffic changes, failures). We compare GraphCC with a state-of-the-art MARL solution for ECN tuning, and observe that our method outperforms the state-of-the-art baseline in all evaluation scenarios, with improvements up to 20% in average Flow Completion Time, similar mean throughput (within 1%), and significant reductions in buffer occupancy (38.0–85.7%).

Cellular networks rely on handovers (HOs) as a fundamental element to enable seamless connectivity for mobile users. A comprehensive analysis of HOs can be achieved through data from Mobile Network Operators (MNOs); however, the vast majority of studies employ data from measurement campaigns within confined areas and with limited end-user devices, thereby providing only a partial view of HOs. This paper presents the first countrywide analysis of HO performance, from the perspective of a top-tier MNO in a European country. We collect traffic from approximately 40M users for 4 weeks and study the impact of the radio access technologies (RATs), device types, and manufacturers on HOs across the country. We characterize the geo-temporal dynamics of horizontal (intra-RAT) and vertical (inter-RATs) HOs, at the district level and at millisecond granularity, and leverage open datasets from the country's official census office to associate our findings with the population. We further delve into the frequency, duration, and causes of HO failures, and model them using statistical tools. Our study offers unique insights into mobility management, highlighting the heterogeneity of the network and devices, and their effect on HOs.

Network models are an essential block of modern networks. For example, they are widely used in network planning and optimization. However, as networks increase in scale and complexity, some models present limitations, such as the assumption of Markovian traffic in queuing theory models, or the high computational cost of network simulators. Recent advances in machine learning, such as Graph Neural Networks (GNN), are enabling a new generation of network models that are data-driven and can learn complex non-linear behaviors. In this paper, we present RouteNet-Fermi, a custom GNN model that shares the same goals as Queuing Theory, while being considerably more accurate in the presence of realistic traffic models. The proposed model predicts accurately the delay, jitter, and packet loss of a network. We have tested RouteNet-Fermi in networks of increasing size (up to 300 nodes), including samples with mixed traffic profiles — e.g., with complex non-Markovian models — and arbitrary routing and queue scheduling configurations. Our experimental results show that RouteNet-Fermi achieves similar accuracy as computationally-expensive packet-level simulators and scales accurately to larger networks. Our model produces delay estimates with a mean relative error of 6.24% when applied to a test dataset of 1,000 samples, including network topologies one order of magnitude larger than those seen during training. Finally, we have also evaluated RouteNet-Fermi with measurements from a physical testbed and packet traces from a real-life network.

Emerging applications such as the metaverse, telesurgery or cloud computing require increasingly complex operational demands on networks (e.g., ultra-reliable low latency). Likewise, the ever-faster traffic dynamics will demand network control mechanisms that can operate at short timescales (e.g., sub-minute). In this context, Traffic Engineering (TE) is a key component to efficiently control network traffic according to some performance goals (e.g., minimize network congestion).This paper presents Routing By Backprop (RBB), a novel TE method based on Graph Neural Networks (GNN) and differentiable programming. Thanks to its internal GNN model, RBB builds an end-to-end differentiable function of the target TE problem (MinMaxLoad). This enables fast TE optimization via gradient descent. In our evaluation, we show the potential of RBB to optimize OSPF-based routing (≈25% of improvement with respect to default OSPF configurations). Moreover, we test the potential of RBB as an initializer of computationally-intensive TE solvers. The experimental results show promising prospects for accelerating this type of solvers and achieving efficient online TE optimization.

Network modeling is a fundamental tool in network research, design, and operation. Arguably the most popular method for modeling is Queuing Theory (QT). Its main limitation is that it imposes strong assumptions on the packet arrival process, which typically do not hold in real networks. In the field of Deep Learning, Graph Neural Networks (GNN) have emerged as a new technique to build data-driven models that can learn complex and non-linear behavior. In this paper, we present RouteNet-Erlang, a pioneering GNN architecture designed to model computer networks. RouteNet-Erlang supports complex traffic models, multi-queue scheduling policies, routing policies and can provide accurate estimates in networks not seen in the training phase. We benchmark RouteNet-Erlang against a state-of-the-art QT model, and our results show that it outperforms QT in all the network scenarios.

Graph neural networks (GNN) have recently been proposed as a technique for accurate and cost-efficient network modeling. As an example, the GNN-based model RouteNet has shown potential for network performance evaluation, being the first-of-its-kind Machine-Learning-based model with generalization capabilities to other networks and configurations unseen during training.In this paper we assess the generalization limits of RouteNet, by analyzing how different network parameters affect the accuracy of this model. To this end, we systematically evaluate the accuracy of RouteNet under modifications of properties of the network and the traffic, such as the topology size, link capacities, the packet size distribution, and the network congestion level. We determine that, while this GNN model is robust to changes in the structure of its input graph, the quality of the estimates degrades considerably, when the distributions of the predicted values of the evaluation data differ from the training (e.g., end-to-end delays). As a result, we argue that to achieve practical GNN-based solutions for network modeling, new methods are needed that can, for example, cope with traffic loads and network sizes that are significantly different than those seen during training.

Network modeling is a critical component of Quality of Service (QoS) optimization. Current networks implement Service Level Agreements (SLA) by careful configuration of both routing and queue scheduling policies. However, existing modeling techniques are not able to produce accurate estimates of relevant SLA metrics, such as delay or jitter, in networks with complex QoS-aware queueing policies (e.g., strict priority, Weighted Fair Queueing, Deficit Round Robin). Recently, Graph Neural Networks (GNNs) have become a powerful tool to model networks since they are specifically designed to work with graph-structured data. In this paper, we propose a GNN-based network model able to understand the complex relationship between (i) the queueing policy (scheduling algorithm and queue sizes), (ii) the network topology, (iii) the routing configuration, and (iv) the input traffic matrix. We call our model TwinNet, a Digital Twin that can accurately estimate relevant SLA metrics for network optimization. TwinNet can generalize to its input parameters, operating successfully in topologies, routing, and queueing configurations never seen during training. We evaluate TwinNet over a wide variety of scenarios with synthetic traffic and validate it with real traffic traces. Our results show that TwinNet can provide accurate estimates of end-to-end path delays in 106 unseen real-world topologies, under different queuing configurations with a Mean Absolute Percentage Error (MAPE) of 3.8%, as well as a MAPE of 6.3% error when evaluated with a real testbed. We also showcase the potential of the proposed model for SLA-driven network optimization and what-if analysis.

The digital transformation is pushing the existing network technologies towards new horizons, enabling new applications (e.g., vehicular networks). As a result, the networking community has seen a noticeable increase in the requirements of emerging network applications. One main open challenge is the need to accommodate control systems to highly dynamic network scenarios. Nowadays, existing network optimization technologies do not meet the needed requirements to effectively operate in real time. Some of them are based on hand-crafted heuristics with limited performance and adaptability, while some technologies use optimizers which are often too time-consuming. Recent advances in Deep Reinforcement Learning (DRL) have shown a dramatic improvement in decision-making and automated control problems. Consequently, DRL represents a promising technique to efficiently solve a variety of relevant network optimization problems, such as online routing. In this paper, we explore the use of state-of-the-art DRL technologies for real-time routing optimization and outline some relevant open challenges to achieve production-ready DRL-based solutions.

Traffic Engineering (TE) is a basic building block of the Internet. In this paper, we analyze whether modern Machine Learning (ML) methods are ready to be used for TE optimization. We address this open question through a comparative analysis between the state of the art in ML and the state of the art in TE. To this end, we first present a novel distributed system for TE that leverages the latest advancements in ML. Our system implements a novel architecture that combines Multi-Agent Reinforcement Learning (MARL) and Graph Neural Networks (GNN) to minimize network congestion. In our evaluation, we compare our MARL+GNN system with DEFO, a network optimizer based on Constraint Programming that represents the state of the art in TE. Our experimental results show that the proposed MARL+GNN solution achieves equivalent performance to DEFO in a wide variety of network scenarios including three real-world network topologies. At the same time, we show that MARL+GNN can achieve significant reductions in execution time (from the scale of minutes with DEFO to a few seconds with our solution).

Network modeling is a key enabler to achieve efficient network operation in future self-driving Software-Defined Networks. However, we still lack functional network models able to produce accurate predictions of Key Performance Indicators (KPI) such as delay, jitter or loss at limited cost. In this paper we propose RouteNet, a novel network model based on Graph Neural Network (GNN) that is able to understand the complex relationship between topology, routing, and input traffic to produce accurate estimates of the per-source/destination per-packet delay distribution and loss. RouteNet leverages the ability of GNNs to learn and model graph-structured information and as a result, our model is able to generalize over arbitrary topologies, routing schemes and traffic intensity. In our evaluation, we show that RouteNet is able to predict accurately the delay distribution (mean delay and jitter) and loss even in topologies, routing and traffic unseen in the training (worst case MRE = 15.4%). Also, we present several use cases where we leverage the KPI predictions of our GNN model to achieve efficient routing optimization and network planning.

Recent advances in Deep Reinforcement Learning (DRL) techniques are providing a dramatic improvement in decision-making and automated control problems. As a result, we are witnessing a growing number of research works that are proposing ways of applying DRL techniques to network-related problems such as routing. However, such proposals failed to achieve good results, often under-performing traditional routing techniques. We argue that successfully applying DRL-based techniques to networking requires finding good representations of the network parameters: feature engineering. DRL agents need to represent both the state (e.g., link utilization) and the action space (e.g., changes to the routing policy). In this paper, we show that existing approaches use straightforward representations that lead to poor performance. We propose a novel representation of the state and action that outperforms existing ones and that is flexible enough to be applied to many networking use-cases. We test our representation in two different scenarios: (i) routing in optical transport networks and (ii) QoS-aware routing in IP networks. Our results show that the DRL agent achieves significantly better performance compared to existing state/action representations.