Adaptive Distributed Architectures for Heterogeneous IoT Ecosystems

The Internet of Things has evolved into a pervasive reality spanning industrial automation, environmental monitoring, smart cities, and healthcare. Yet the contemporary IoT landscape remains deeply fragmented: heterogeneous protocols create isolated ecosystems, mobile devices cannot roam seamlessly across operator boundaries, centralized cloud architectures introduce latency and single points of failure, and the absence of economic incentives undermines voluntary participation in shared infrastructure. This thesis investigates adaptive distributed architectures that address these challenges not by eliminating heterogeneity through standardization, but by designing mechanisms that function correctly in its presence. Six interconnected contributions are presented. First, a gateway-based architecture integrating Stack4Things and the Data eXchange Mediator Synthesizer bridges heterogeneous IoT protocols while exposing unified RESTful interfaces through a dynamic DNS mechanism that ensures public accessibility regardless of NAT and firewall constraints. Second, an API-driven LoRaWAN gateway bridge service enables seamless device roaming across operator boundaries without pre-established agreements, firmware modifications, or Network Server patches, achieving end-to-end latency within industrial IoT bounds and zero gateway-level packet loss. Third, a blockchain-based SLA framework on the Algorand platform automates provider registration, contract negotiation, and payment settlement for multi-stakeholder IoT infrastructures, enforcing forwarding-level QoS guarantees without centralized coordination. Fourth, a WebRTC-based P2P overlay network operating at Layer~2 enables direct communication among IoT devices across heterogeneous administrative domains, achieving order-of-magnitude latency reductions compared to cloud-mediated architectures while remaining transparent to all higher-layer protocols and legacy applications. Fifth, a topology optimization framework for resource-constrained scenarios introduces a device categorization model and multi-objective cost formulation, extended to dynamic mobile deployments through graph-based trajectory prediction for UAV fleet coordination. Sixth, an empirical comparison of FedAvg and FedProx on the SLICES research infrastructure demonstrates that proximal regularization and partial-update tolerance make FedProx robustly superior whenever system or statistical heterogeneity is present, providing principled algorithm selection guidance for federated deployments over heterogeneous edge infrastructures. Taken together, these contributions demonstrate that IoT heterogeneity can be managed through adaptive distributed architectures that embrace diversity rather than suppress it, and that edge-based infrastructures can support practical distributed applications while maintaining data locality, respecting privacy constraints, and operating independently of centralized cloud services.