UC Irvine

This dissertation develops a comprehensive, application-aware approach to optimizing the full lifecycle of Internet of Things (IoT) systems, from planning to operation. The core thesis is that a holistic approach must address three distinct timescales: long-term strategic planning, short-term event-driven deployment, and real-time network operation. This work presents a novel, integrated suite of frameworks to address challenges from planning to operation. First, to address optimized planning of permanent urban IoT infrastructure, we introduce SmartParcels, a human-in-the-loop, cross-layer framework that balances deployment and operation costs, device reuse, and application sensing needs. We next address the issues of rapidly deployed temporary IoT installations using smart firefighting applications as a driving usecase. Here we integrate a model-driven methodology to capture the evolution of dynamic physical phenomena (e.g., fire spread in prescribed burns) using a novel criticality metric to guide sensor placement and maximize anomaly detection. Finally, to address real-time operational dynamics associated with data delivery in extreme situations, we propose TINMAN, an interference-resilient framework for multinetwork communication. TINMAN uses a novel machine learning approach to predict link delays by distinctly quantifying the impact of exogenous traffic and ambient signal interference. A centralized controller leverages these predictions to construct optimal paths for delivery of time-critical data. Evaluations using real-world deployments and data with high-fidelity simulations, including those developed by experts, demonstrate that this three-part approach significantly improves cost-performance, anomaly detection rates, and data-delivery timeliness.