Modern edge–cloud infrastructures are composed of highly distributed components whose interactions evolve dynamically over time. Observability tools typically provide fragmented views of system metrics, logs, and traces, making it difficult to understand relationships between resources and their communication patterns. This paper presents an architecture that combines a data acquisition layer and a dynamic graph modelling approach to improve the observability of distributed systems. The proposed system is composed of two main components: the Telemetry Data Collector and Monitoring Engine (TDCME) and the Dynamic Graph Modeller (DGM). TDCME is responsible for collecting heterogeneous data from cloud-native infrastructures, including Kubernetes resources and network flows. DGM transforms the collected information into a dynamic graph representation where infrastructure entities and their relationships are continuously updated and visualized. The resulting graph enables real-time exploration of system dependencies and communication patterns. The proposed approach facilitates improved understanding of distributed edge–cloud environments and provides an interactive mechanism for analyzing infrastructure behavior.

Swarm Computing is getting significant traction recently, as means to exploit the immense power of cloud computing continuum resources. Modern data-intensive and IoT applications, especially those in the domain of artificial intelligence, require efficient and continuous management support with respect to where their components are deployed, how close to their required data sources and/or to their end-users and how they should scale as a response to significant fluctuations in their incoming workload or due to sudden failures of their hosting infrastructure. We claim that the hosting infrastructure should include any available cloud, fog, and edge nodes in vicinity to their data sources, while the applications deployment or reconfigurations, over such nodes, should involve decentralized orchestration capabilities to cope with the dynamicity of the Cloud-to-Edge continuum. Several recent efforts introduce such decentralized orchestration platforms with notably appropriate capabilities that cope with dynamic settings. In this work, we focus on one of these platforms, called Swarmchestrate, focusing on the trustworthy management of any data artefacts that are bound to be exchanged in terms of such management procedures. The limitations of centralised solutions are addressed by going a step forward presenting a holistic vertical trust layer which will benefit swarm computing environments. Specifically, our proposal introduces the Swarmchestrate trust model that will be implemented through a well-integrated solution comprising blockchain-enabled decentralized identifiers (DIDs) that authenticate and authorize the access to data and knowledge artefacts across a peer-to-peer network of data/knowledge agents. In this approach, the DIDs used can be significantly affected by real-time telemetry data that are used as one of the means to verify the reputation of the involved entities (i.e. decentralised agents, users, application components).

The increasing proliferation of distributed computing infrastructures spanning Edge, Fog, and Cloud environments introduces new challenges for application development, deployment, and execution. These environments are characterized by dynamic conditions such as fluctuating resource availability, variable network latency and bandwidth, changing workloads, node mobility and churn, and heterogeneous hardware and software capabilities. Such variability can significantly impact application performance, reliability, and quality of service at runtime. Consequently, adaptation is required to enable applications to respond to these changes, for example by dynamically scaling, reconfiguring components, or migrating workloads across the continuum to maintain desired operational objectives. However, existing approaches to adaptation remain largely infrastructure-driven, limiting the ability of applications to respond to dynamic runtime conditions. This paper presents an Application Controller (AC) framework that enables applications to exhibit more responsive and intelligent behaviours, while pursuing their performance goals, energy or cost parameters or policy constraints, in the computing continuum. The AC is designed as a reusable open source library of reconfigurable functions that can be directly embedded within applications. The use of AC allows applications to monitor their runtime behaviour, evaluate application-specific performance goals or conformance requirements, and proactively implement adaptation actions to optimise the application behaviour. The adaptation actions can range from load balancing and elasticity to resource management and adaptive scheduling. The AC is delivered as part of an Application Programming Model (APM), which provides a much broader modular framework and tooling for building adaptive applications that can transition from passive entities managed by infrastructure to active participants in their own optimisation. The paper details the design principles, architecture, and operational workflow of the AC, and discusses its role in enabling responsive, policy-driven adaptation across the Edge-to-Cloud continuum.

Task allocation across the cloud--edge continuum often uses multi-criteria optimization balancing latency, energy efficiency, and reliability. However, such approaches generally assume that for a given configuration, all criteria can be computed quickly enough to be used directly in a single reward function during runtime. Not all criteria fit this, for example performing a security and compliance risk assessment of the proposed configuration is much slower than the rest, because it considers interactions between components, environment and even user input. This problem of radically different latencies for criteria estimation arises in many different contexts, and existing solutions delay results until the slowest latency criteria has been calculated, which is not realistic in many applications. To address this problem, we propose the use of a novel three-agent architecture. The first agent performs dynamic optimization of the computational task allocation using online DRL and GNN. A second agent, called Security Risk Modeler (SRM), performs a holistic risk assessment for proposed configurations at "its own time" which is not compatible with the optimization speed of the first agent. The third agent learns to approximate SRM risk assessment results in real time so that the agent optimizing task allocation can perform at the expected operational speed. The architecture follows agentic AI principles because it decomposes the task allocation function into specialized interacting agents with distinct roles, responsibilities and timelines. The paper proposes the conceptual model and architectural design of the novel solution plus pseudo-code for future experimentation. Overall, the paper contributes to the future vision of trustworthy, governable, and risk-aware AI systems.

The growing digitalization of advanced industries has driven to the need for more effective and trustworthy data integration and sharing among disparate organizations and their systems. Data Space has emerged as a comprehensive approach to address this, enabling data access respecting security, compliance, governance, interoperability, and sovereignty through a federated and integrated strategy. Current proposals and implementations have essentially applied data spaces in interorganizational scenarios. However, modern manufacturing environments also require more robust and holistic infrastructures in a way that enterprise systems can properly have access to cyber-physical systems (CPS) data while data ownership and sovereignty are granted. This paper proposes a novel model of data spaces usage in a way to support manufacturing intraorganizational needs, at the shop floor level. Grounded on the GAIA-X, IDSA, and RAMI 4.0 / AAS reference architectures, the model allows authorized systems to access shareable CPS data in real time for their different purposes, where the data space acts as an intermediary and federated data sharing management system. A software prototype has been implemented to showcase the proposed model as a proof-of-concept, demonstrating its architectural feasibility and potential.