Satellite offloading is a critical issue in the Internet of Things (IoT) edge intelligence environment. In this work, we present a novel neurosymbolic framework for computation offloading decisions in satellite-enabled IoT edge intelligence scenarios. By combining the forecasting capabilities of time-series foundation models with the transparency of rule-based reasoning, our approach enables data-efficient and inherently explainable decision making under uncertainty. Specifically, we use TimeGPT to predict future throughput quality and satellite CPU load, which are then processed through a fuzzy logic controller to derive context-aware offloading decisions with transparent rationale. The results show effective forecast accuracy, high decision robustness, and improved explainability when compared to traditional reinforcement learning-based approaches that require task-specific training.

The growing size and complexity of Large Language Models (LLMs) pose major challenges for deployment in edge environments with limited computational and communication resources. To address this, we propose a novel distributed framework for allocating LLM layers across heterogeneous edge nodes. By modeling the problem as a two-sided matching game, where layers and nodes rank each other based on processing delay and inter-node transmission latency, our approach achieves stable and low-latency allocations without centralized coordination. In particular, we first design a lightweight matching algorithm that accounts for pipeline dependencies and mitigates idle periods during sequential inference. We then extend the proposed framework to support multi-tenant scenarios, where multiple LLMs compete for shared edge resources. Extensive simulations show that our approach achieves up to a 10% reduction in inference latency compared to the Kolkata Game baseline. We also validate the proposed framework with a real-world testbed based on LLaMA-7B under realistic conditions.

Internet of things (IoT) applications require up-todate information about the system conditions. This can often be provided from multiple alternative sources that sense the environment, but act without centralized coordination. In this paper, we consider a scenario where multiple sources can provide information for a number of tasks of the IoT application, assuming that the information content of multiple sources is generally redundant, yet one single source is generally insufficient for all tasks. In so doing, we seek for the minimization of the age of federated information (AoFI), a metric describing the age of information from multiple sources, considering that the epochs of successful updates are only those where all tasks are covered. At the same time, we would like to contain the number of active sources for cost reasons. To this end, we tackle the problem through a game-theoretic approach, where individual sources act as players minimizing a linear combination of AoFI and activation cost. We prove that this framework identifies efficient Nash equilibria very close to the optimum performance. However, the latter can only be achieved through centralized control, whereas the former allows for distributed implementation, which is key in IoT scenarios.

Apache Kafka is a key component in event-driven and microservice architectures relying on distributed publishsubscribe messaging for scalable and fault-tolerant streaming of real-time data. To reduce distribution overhead, messages are buffered and dispatched to the broker when either a maximum batch size N is reached or a timeout T expires, enabling control on the trade-off between high throughput and low latency. However, this trade-off has been explored only through empirical studies, referred to specific system deployments and not suited for runtime adaptation to variable workload conditions. We provide an analytical characterization of the arrival process induced by Kafka batching policy under Poisson arrivals. The analysis develops on the observation that the time for buffering a full batch and the size of a batch dispatched at expiration of the timeout follow truncated Erlang and Poisson distributions, respectively. Leveraging this insight, we derive closed forms for quantities that characterize the arrival process, and we propose a method for efficient simulation of the process embedded at dispatching times. To support practical implementation, we evaluate solutions for drawing samples from an Erlang distribution provided by NumPy, PyTorch, and R, and we also propose a novel approach based on rejection sampling with proposal function in the family of Kumaraswamy distributions with automated optimization of parameters with respect to the number of phases. Numerical experimentation shows that: (i) aggregated simulation enabled by the analytical formulation is insensitive to the value of the batch size, and it definitely outperforms fine-grained simulation of individual message arrivals; (ii) the best efficiency is obtained with the NumPy implementation of Erlang, with promising results of the novel approach based on Kumaraswamy, which achieves results comparable to PyTorch and better than $\mathbf{R}$.

Artificial intelligence (AI)-native edge networks are promising solutions to seamlessly integrate AI into modern network architectures, promoting intelligent and hyper-flexible behavior in self-adaptation and reconfiguration of hybrid networks. AI-native edge nodes have to promptly react to any change in network conditions and to be linked to Internet of Everything (IoE) deployed in etherogeneous communication domains. This paper deals with a next-generation programmable edge node capable of being employed in different domains in a unified and flexible manner. With the aim to achieve a low re-configuration overhead in such herogeneous contexts, this paper proposes the integration of a generative-AI module within an edge node exploiting foundation models for efficient general-purpose time-series prediction, without involving overhead and costs due to models trained from scratch and overcoming data scarcity. This permits to manage IoE networks deployed in both homogeneous and heterogeneous domains, i.e., aqua, ground and air, autonomously, without the need for adjustments from the outside. As foundation models, we focused on Chronos and TimesFM, in both the zero-shot and fine-tuning learning paradigms. Finally, performance results in terms of prediction accuracy, training, and inference time are provided and compared with those achieved by the state-of-the-art recurrent neural networks trained from scratch and baseline alternatives. The obtained results corroborate the potential of foundation models as key enablers of native AI networks, achieving good accuracy despite the absence of a training phase (i.e., in zero-shot mode) or with limited training (fine-tuning), w.r.t. alternatives.

Current edge computing (EC) solutions face the significant challenge of limited computational capacities. Effectively allocating resources and controlling the system to ensure task timeliness remains an open problem. The age of information (AoI) is a metric to measure the freshness of information that circulates in a system. While the AoI is primarily influenced by packet generation rate, transmission latency, and queuing delays, the processing time becomes notably significant when dealing with Internet-of-Things (IoT) computationally intensive tasks. Such IoT applications necessitate processing before embedded information can emerge and status can be acquired. This paper proposes a combined system control and resource assignment policy, in new-generation EC environments, where edge nodes have limited capacity and task flows are computationally intensive. The objective is to assign task flows to dedicated resource capacity, minimizing the worst AoI experienced by flows. For this purpose, three problem formulations for the flow-resource assignment are considered: i) the assignment with fixed arrival and service processes; ii) the service process control problem; iii) the arrival process control problem. For each problem formulated, a matching game with externalities is designed, and preference lists are built considering the mean AoI of an M/G/1 system, here exploited as reference model to represent each computation partition. The stability of matching games proposed is investigated, and experimental results are presented to highlight the validity of the matching approaches, providing critical discussion about the performance impact of the three problems addressed, also compared with a reservoir learning approach. The proposed matching algorithm surpasses the state-of-the-art Deferred Acceptance method by achieving a lower maximum AoI, thereby meeting the optimization objective. It also demonstrates improved performance over the data-driven approach. While comparable maximum AoI values can be attained with sufficiently large training datasets, the proposed algorithm consistently yields superior results.

The ever-increasing complexity of modern software architectures has exacerbated the need for advanced software tools able to track software execution traces to improve software reliability. In this paper, we present OREO, a tool for offline and run-time monitoring and fault localization. The tool implements a novel method enabling to trace software executions to discover the run-time status, dependencies, and interactions among software components. OREO is based on a timeline extractor, i.e., an abstraction of component lifecycles and their interactions. The timeline extractor enables the tool to perform a runtime health state examination of the software under analysis. The profiler is then used to analyze the error propagation originated during the running states among software components. In so doing, the possible fault-error-failure chains are identified. To showcase the capabilities of OREO and its flexibility, we report the execution of the tool on three software projects of different nature, sizes, and architectures. The analysis results in the localization of fault-error-failure chains and safe components of the three software projects. A discussion of the versatility, scalability, and applicability of the proposed tool to a rich variety of application contexts is provided.

Effective exploitation of Machine Learning solutions in the Digital Twin (DT) paradigm may largely benefit from Federated Analytics (FA) approaches, to mitigate data scarcity, by merging distributed data, and heterogeneity while limiting communication overhead and exchange of sensitive raw data. In the DT paradigm, federated schemes find a native collocation in the conceptual association between Digital Twin Prototype (DTP) of a class and Digital Twin Instance (DTI) of individual products. We propose the application of the democratized learning (Dem-AI) scheme to provide a scalable solution for multi-level hierarchical integration of data owned by a multiplicity of distributed DTs, and we showcase its application in a failure prediction scenario. The proposed model integration scheme preserves the inherent cohesive relationships between generalization and specialization (or personalization) capabilities of the DTP model and the DTI model, respectively. Based on the model acquired, DTs monitor the system behavior and forecast failure occurrences. Experimental analysis has been conducted to thoroughly investigate the performance of the Dem-AI failure prediction framework designed, considering different levels of model specialization and public dataset.

The Internet of Things (IoT) paradigm is assumed to be a major component in the present and future Internet, with forecasts claiming a humongous number of devices connected in the near future, and applications fields spanning from agriculture, to healthcare. Despite this, the standardization efforts have not yet resulted in widely adopted standards, and the market is fragmented into multiple solutions both at physical and communication protocol levels. Moreover, IoT systems exacerbate the usual test bed limitations, e.g., scalability (very large number of devices), hardware compatibility, space, and price. Due to the above problems, simulation tools become an extremely interesting tool for studying IoT systems both for academia (new algorithms), standardization (new protocols), and industry (what-if analysis). In this paper we will discuss what are the most relevant features and models that a simulation tool like ns-3 should prioritize to enable the above-mentioned needs from academia, standardization, and industry, and if they are achievable in the short, medium, or long term.

Nowadays, an ambitious target of the next generation networks is to develop intelligent overarching space-air-ground-aqua computing systems, in order to provide a smart ecosystem able to efficiently operate computation in heterogeneous domains. In particular, in such a context, the underwater environment requires a special attention, since it is recognized as the most challenging domain, due to channel impairments and adverse propagation conditions. This paper proposes a self-intelligent system able to efficiently perform underwater environment monitoring or underwater survey of critical infrastructure, by resorting to the use of the semantic communication paradigm to lower the impairments due to the underwater channel propagation conditions. In particular, in our case, images sent by underwater devices are collected by shore small base stations (SSBSs) to form their training dataset to take part in a federated learning process with a ground base station. In particular, the paper considers a semantic communication scheme based on a deep-convolution neural networks encoder-decoder architecture for an efficient exploitation of the data transmission from underwater devices to the linked SSBSs. Performance analysis is provided to show the better behavior of the proposed system in comparison with the conventional alternative that does not involve the use of the semantic communications approach. Finally, a specific performance evaluation analysis is devoted to the investigation of the convergence behavior of the proposed federated learning procedure in reference to the cross ground-aqua system considered in order to highlight its advantages with respect to a classical implementation.

Nowadays, the functional integration of Unmanned Aerial Vehicles (UAVs) as flying computing nodes with terrestrial networks is rapidly emerging as a promising and viable solution to enhance performance or lower drawbacks arising from unpredictable traffic load congestion occurrences. In particular, this paper considers a UAV-Aided Multiple Access Edge Computing system, in which heterogeneous traffic flows with different quality of service constraints, have to be offloaded on processing nodes consisting of terrestrial and flying edge computing nodes. Towards this goal, the paper proposes a matching algorithm to perform an efficient offloading strategy. In particular, the proposed matching algorithm provides decisions on the basis of per-flow end-to-end delay bounds formulated by resorting to the combined application of stochastic network calculus and martingale envelopes theory. Furthermore, matching stability has been theoretically discussed. Numerical results highlight the validity of the proposed stochastic framework in terms of both reliability, i.e., the probability with which the per-flow end-to-end delay is lower than the corresponding deadline, and its ability to fit the actual network behavior. For comparison purposes, the Boole bound is formulated, and a greedy algorithm is developed to compare the matching strategy designed.

This article deals with an artificial intelligence (AI) framework to support Internet-of-everything (IoE) applications over sixth-generation wireless (6G) networks. An integrated IoE-Edge Intelligence ecosystem is designed to effectively face the problems of Virtual Machines (VMs) placement based on their popularity, computation offloading optimization, and system reliability improvement predicting compute nodes faults. The main objective of the article is to increase performance in terms of minimization of worst end-to-end (e2e) delay, percentage of requests in outage, and the enhancement of reliability. The article focuses on the following main issues: (i) proposal of a channel-aware federated learning (FL) approach to forecast the popularity of the VMs required by IoE devices; (ii) use of an AI-based channel conditions forecasting module at the benefits of the FL process; (iii) development of a suitable VMs placement on the basis of their popularity and of an efficient tasks allocation technique based on a modified version of the auction theory (AT) and a proper matching game; (iv) enhancement of the system reliability by an echo-state-network (ESN), located on each computation node and running in the background to predict failures and anticipate tasks migration. Numerical results validate the effectiveness of the proposed strategy for IoE applications over 6G networks.

The forthcoming sixth-generation (6G) network technology will trigger tremendous changes across several application scenarios and even in the way we work, communicate, and, more in general, organize our everyday life. In particular, in such a new digital era, virtual reality (VR) technology will play a fundamental role, carrying with it numerous challenges as regards the increasing demand of high rate, highly reliable, and low-latency communications, especially in reference to Internet-of-Things environments. In this sense, an effective end-to-end (e2e) delay analysis becomes imperative to pursue an efficient design of the foreseen VR services. Toward this goal, this article proposes the e2e delay investigation through the martingale theoretical bound and considering the stochastic network calculus principles for its formulation. The performance discussion focuses on the achieved communications reliability, comparing the analytical predictions obtained via the martingale bound with the simulation results, and the Markov queuing theory-based alternative. The presented results confirm the validity of the proposed approach, exhibiting noticeable closeness between the theoretical bound and the simulation outcomes.

This paper analyzes the end-to-end delay performance in an edge-computing scenario where a set of Internet of Things devices (IoTDs) access the computation facilities of an Edge Node by means of a 5G based network. In particular, the paper deals with a two power levels slotted Aloha non-orthogonal-multiple-access (NOMA) scheme and formulates a stochastic end-to-end delay bound, in terms of complementary cumulative probability distribution, by resorting to the application of the martingale theory. In order to validate the proposed analysis, the paper proposes comparisons between the achieved analytical predictions and actual values derived by resorting to extensive computer simulations. Furthermore, the well known Boole bound has been formulated and compared with the proposed Martingale approach to highlight the better behavior of the proposed solution.

Nowadays, industrial Internet of Things (IIoTs) has gained attention as an emerging application area of the Internet of Things paradigm to improve efficiency and reliability in a wide class of manufacture processes. This article focuses on a combined edge-cloud computing architecture for IIoT systems, and addresses the minimization of both the mean system response time and the number of requests dropping due to a response time greater than their time deadline. The problem is formulated as a matching game with externalities between the edge computing servers, and the applications themselves. The proposed strategy adopts a discard policy with the aim at favoring the IIoT-devices requests with a not-expired deadline. Moreover, a suitable discussion of the matching stability is proposed. Finally, the strategy's validity has been confirmed by the system performance expressed in terms of mean and worst system response time and outage probability, in comparison to methods recently proposed in the literature.

Recently, Edge Computing systems have emerged as a suitable solution to provide fast data processing with low latency to mobile users. In particular, in the case of real time computation demands, the users mobility often plays a very central role, requiring highly responsive information processing and interpretation of the data, retrieved by environment, in order to apply proper strategies for decision making. This paper proposes an approach based on the queuing theory in order to derive the performance of an Edge Computing systems in the case of real time applications demands and user mobility. In particular, on the basis of some simplifying assumptions, the proposed analytical method allows to identify the minimum (i.e., optimal) number of central process units to be allocated to an edge computing node in order to meet specific application constraints. The validity of our approach is then validated by providing performance comparisons between the obtained analytical predictions and simulation results derived by assuming actual application conditions.

Chaos theory constitutes a promising and powerful tool to address forecasting problems of nonlinear time series, since it catches the dynamical and geometrical structure of very complex systems, ensuring a superior accuracy on the predicted results in comparison to classical approaches, and lowering the complexity typical of the deep learning ones. This paper applies the nonlinear chaos theory principles to the passenger demand forecasting problem. The proposed scheme processes and analyzes the big data of passengers requests collected during one month in the city of Chengdu, China. In order to predict passenger demand behavior, the chaotic trend of the time series has been identified through the largest Lyapunov exponent research. Then, the phase space reconstruction has been pursued, through the detection of the suitable embedding dimension and time delay, to improve forecasting accuracy and avoid information redundancy. A combined local and global predictive method has been proposed, and the validity of the approach is confirmed by comparison with two state-of-art forecasting methods, the auto-regressive and the auto-regressive moving-average with exogenous input. System performance is evaluated in terms of mean squared error and mean percent forecasting error.

The explosive growth of mobile traffic needs that the fifth generation of wireless systems supports a thousand-fold increase of network capacity, thanks to new paradigms such as network densification. However, in dense networks interference can be a limiting factor, and must be suitably managed. In this paper multiple cells coordinate to mitigate the mutual interference: cross-tier interference is limited by means of the traditional enhanced Inter Cell Interference Coordination approach, while a new co-tier interference management strategy based on Matching Theory is proposed. The objective is to provide a stable matching between the set of small cells and the set of available subframes left almost blank by the macrocell and used by the small cells to communicate with their most vulnerable users. The proposed solution aims at improving the throughput of most critical users with an affordable computational complexity. The effectiveness of the proposed scheme has been verified in comparison with benchmark methods.

This paper proposes two virtual functions (VFs) placement approaches in a Fog domain. The considered solutions formulate a matching game with externalities, aiming at minimizing both the worst application completion time and the number of applications in outage, i.e., the number of applications with an overall completion time greater than a given deadline. The first proposed matching game is established between the VFs set and the fog nodes (FNs) set by taking into account the ordered sequence of services (i.e., chain) requested by each application. Conversely, the second proposed method overlooks the applications service chain structure in formulating the VF placement problem, with the aim at lowering the computation complexity without loosing the performance. Furthermore, in order to complete our analysis, the stability of the reached matchings has been theoretically proved for both the proposed solutions. Finally, performance comparisons of the proposed matching theory approaches with different alternatives are provided to highlight the superior performance of the proposed methods.

Fog Computing (FC) is an emerging paradigm that extends cloud computing toward the edge of the network. In particular, FC refers to a distributed computing infrastructure confined on a limited geographical area within which some Internet of Things applications/services run directly at the network edge on smart devices having computing, storage, and network connectivity, named fog nodes (FNs), with the goal of improving efficiency and reducing the amount of data that needs to be sent to the Cloud for massive data processing, analysis, and storage. This paper proposes an efficient strategy to offload computationally intensive tasks from end-user devices to FNs. The computation offload problem is formulated here as a matching game withexternalities, with the aim of minimizing the worst case service time by taking into account both computational and communications costs. In particular, this paper proposes a strategy based on the deferred acceptance algorithm to achieve the efficient allocation in a distributed mode and ensuring stability over the matching outcome. The performance of the proposed method is evaluated by resorting to computer simulations in terms of worst total completion time, mean waiting, and mean total completion time per task. Moreover, with the aim of highlighting the advantages of the proposed method, performance comparisons with different alternatives are also presented and critically discussed. Finally, a fairness analysis of the proposed allocation strategy is also provided on the basis of the evaluation of the Jain’s index.

This paper analyzes the resource allocation problem in the LTE-Unlicensed spectrum to increase performance of the unlicensed systems. The problem is modeled with a matching game approach and the proposed resource allocation algorithm consists of a heuristic based on the usage of preference lists, one for each user equipment (UE) and on unlicensed bands, to determine a many-to-many matching between UEs and component carriers (CCs), aiming at maximizing the unlicensed overall network sum rate. Hence the proposed algorithm realizes a mapping between user equipments and unlicensed channels, also considering the carrier aggregation (CA) mechanism, which is typical of LTE-U. In the simulation evaluation, we compare the proposed algorithm with the Hungarian algorithm w.r.t. the system throughput and the robustness to perturbed input data. Both algorithms have almost equivalent performances, but the Hungarian algorithm presents a higher computational complexity than the proposed one.