THE evolution of modern communication networks is increasingly driven by intelligent and data-intensive applications deployed over heterogeneous, distributed, and highly dynamic infrastructures. Modern networks must integrate edge and cloud computing, terrestrial and nonterrestrial connectivity, and virtualized network functions under strict constraints related to resource availability, energy efficiency, security, and resilience. This architectural shift introduces fundamental orchestration challenges, as traditional control and optimization approaches are no longer sufficient to cope with unstable connectivity, heterogeneous resources, and diverse service requirements. Emerging paradigms such as Software-Defined Wide Area Network (SD-WAN)–orchestrated edge–cloud systems, satelliteassisted connectivity, and virtualized 5G Radio Access Networks (RAN) further increase system complexity by spanning multiple administrative domains and technology layers. In such environments, orchestration must continuously adapt to dynamic conditions while jointly coordinating computation, communication, and learning processes in a scalable and resourceefficient manner. This thesis investigates resource trading and orchestration in secure edge–cloud networks from a unified perspective aligned with modern Artificial Intelligence (AI)-enabled network architectures. The research explores how Machine Learning (ML) and Federated Learning (FL) can be embedded into network control loops to enable adaptive decision-making under heterogeneous resource constraints, dynamic connectivity conditions, and multidomain operational requirements. The first part focuses on resource-aware model aggregation and decentralized learning strategies for FL in SD-WANenabled Multi-access Edge Computing (MEC) environments, introducing hierarchical and distributed aggregation mechanisms that improve robustness and scalability under limited wide area connectivity. The scope is then extended to heterogeneous infrastructures through satellite-assisted Space–Ground Integrated Networks (SGIN), where predictive and learning-driven traffic steering enables proactive adaptation to highly dynamic topologies. To support heterogeneous services and multioperator deployments, economically grounded network slicing frameworks are developed, combining hierarchical auction mechanisms with regionaware resource allocation to balance efficiency, fairness, and operational cost. Security and resilience are further addressed through the study of Quantum Key Distribution (QKD) networks, where attack-aware routing metrics and progressive recovery mechanisms are proposed to jointly optimize survivability and resource utilization. Overall, this thesis demonstrates that future AI-based communication infrastructures require unified orchestration frameworks that combine intelligence, economic resource trading, multi-domain coordination, and quantumsecure communication. The proposed methodologies contribute scalable design principles for next-generation edge–cloud ecosystems operating under heterogeneous, dynamic, and security-critical conditions.
L’Evoluzione delle moderne reti di comunicazione è sempre più guidata da applicazioni intelligenti e ad alta intensità di dati, implementate su infrastrutture eterogenee, distribuite e altamente dinamiche. Le reti moderne devono integrare edge computing e cloud computing, connettività terrestre e non terrestre e funzioni di rete virtualizzate, nel rispetto di rigorosi vincoli legati alla disponibilità delle risorse, all’efficienza energetica, alla sicurezza e alla resilienza. Questo cambiamento architetturale introduce sfide fondamentali in termini di orchestrazione, poiché i tradizionali approcci di controllo e ottimizzazione non sono più sufficienti a far fronte a connettività instabile, risorse eterogenee e requisiti di servizio diversificati. Paradigmi emergenti come i sistemi edge-cloud orchestrati tramite Software-Defined Wide Area Network (SD-WAN), la connettività satellitare e le reti di accesso radio 5G virtualizzate (RAN) aumentano ulteriormente la complessità del sistema, estendendosi a più domini amministrativi e livelli tecnologici. In tali ambienti, l’orchestrazione deve adattarsi costantemente alle condizioni dinamiche, coordinando congiuntamente i processi di elaborazione, comunicazione e apprendimento in modo scalabile ed efficiente in termini di risorse. Questa tesi analizza lo scambio e l’orchestrazione delle risorse in reti edge-cloud sicure da una prospettiva unificata allineata alle moderne architetture di rete basate sull’intelligenza artificiale. La ricerca esplora come il Machine Learning (ML) e il Federated Learning (FL) possano essere integrati nei circuiti di controllo di rete per consentire un processo decisionale adattivo in presenza di vincoli di risorse eterogenei, condizioni di connettività dinamica e requisiti operativi multidominio. La prima parte si concentra sull’aggregazione di modelli basati sulle risorse e sulle strategie di apprendimento decentralizzato per il Federated Learning in ambienti Multi-access Edge Computing (MEC) abilitati per SD-WAN, introducendo meccanismi di aggregazione gerarchica e distribuita che migliorano la robustezza e la scalabilità in condizioni di connettività geografica limitata. L’ambito viene poi esteso a infrastrutture eterogenee attraverso reti integrate spazio-terra (SGIN) assistite da satellite, dove la gestione predittiva e basata sull’apprendimento consente un adattamento proattivo a topologie altamente dinamiche. Per supportare servizi eterogenei e implementazioni multi-operatore, vengono sviluppati framework di network slicing economicamente sostenibili, che combinano meccanismi di asta gerarchica con un’allocazione delle risorse basata sulla regione per bilanciare efficienza, equità e costi operativi. Sicurezza e resilienza vengono ulteriormente affrontate attraverso lo studio delle reti Quantum Key Distribution (QKD), dove vengono proposte metriche di routing basate sugli attacchi e meccanismi di ripristino progressivo per ottimizzare congiuntamente la sopravvivenza e l’utilizzo delle risorse. Nel complesso, questa tesi dimostra che le future infrastrutture di comunicazione basate sull’intelligenza artificiale richiedono framework di orchestrazione unificati che combinino intelligenza, scambio di risorse economiche, coordinamento multidominio e comunicazione quantisticamente sicura. Le metodologie proposte contribuiscono a principi di progettazione scalabili per ecosistemi edge-cloud di nuova generazione che operano in condizioni eterogenee, dinamiche e critiche per la sicurezza.
Resource trading and orchestration in secure edge-cloud networks distributed over heterogeneous domains
LI, MENGYAO
2025/2026
Abstract
THE evolution of modern communication networks is increasingly driven by intelligent and data-intensive applications deployed over heterogeneous, distributed, and highly dynamic infrastructures. Modern networks must integrate edge and cloud computing, terrestrial and nonterrestrial connectivity, and virtualized network functions under strict constraints related to resource availability, energy efficiency, security, and resilience. This architectural shift introduces fundamental orchestration challenges, as traditional control and optimization approaches are no longer sufficient to cope with unstable connectivity, heterogeneous resources, and diverse service requirements. Emerging paradigms such as Software-Defined Wide Area Network (SD-WAN)–orchestrated edge–cloud systems, satelliteassisted connectivity, and virtualized 5G Radio Access Networks (RAN) further increase system complexity by spanning multiple administrative domains and technology layers. In such environments, orchestration must continuously adapt to dynamic conditions while jointly coordinating computation, communication, and learning processes in a scalable and resourceefficient manner. This thesis investigates resource trading and orchestration in secure edge–cloud networks from a unified perspective aligned with modern Artificial Intelligence (AI)-enabled network architectures. The research explores how Machine Learning (ML) and Federated Learning (FL) can be embedded into network control loops to enable adaptive decision-making under heterogeneous resource constraints, dynamic connectivity conditions, and multidomain operational requirements. The first part focuses on resource-aware model aggregation and decentralized learning strategies for FL in SD-WANenabled Multi-access Edge Computing (MEC) environments, introducing hierarchical and distributed aggregation mechanisms that improve robustness and scalability under limited wide area connectivity. The scope is then extended to heterogeneous infrastructures through satellite-assisted Space–Ground Integrated Networks (SGIN), where predictive and learning-driven traffic steering enables proactive adaptation to highly dynamic topologies. To support heterogeneous services and multioperator deployments, economically grounded network slicing frameworks are developed, combining hierarchical auction mechanisms with regionaware resource allocation to balance efficiency, fairness, and operational cost. Security and resilience are further addressed through the study of Quantum Key Distribution (QKD) networks, where attack-aware routing metrics and progressive recovery mechanisms are proposed to jointly optimize survivability and resource utilization. Overall, this thesis demonstrates that future AI-based communication infrastructures require unified orchestration frameworks that combine intelligence, economic resource trading, multi-domain coordination, and quantumsecure communication. The proposed methodologies contribute scalable design principles for next-generation edge–cloud ecosystems operating under heterogeneous, dynamic, and security-critical conditions.| File | Dimensione | Formato | |
|---|---|---|---|
|
Mengyao_Final_Thesis_final.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
22.13 MB
Formato
Adobe PDF
|
22.13 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/256877