Application Aware Data Management in Edge Computing

Abstract

Edge computing confronts a massive data surge from IoT devices, crucial for emerging low-latency applications like AR/VR, autonomous vehicles, and real-time health-care. However, the edge’s inherent limitations, scarce resources, network instability and system heterogeneity, mean that traditional cloud-centric data management strategies cannot be effectively applied ‘out of the box’, necessitating edge-specific approaches. This dissertation posits that application awareness is fundamental to overcoming these challenges, requiring strategies adapted to the unique characteristics and goals of edge applications. We introduce and evaluate four complementary techniques: 1) Cargo, a fault-tolerant, performance-aware storage layer integrated into the Armada edge framework, enabling stateful applications on volatile resources. 2) A comparative analysis of data placement strategies, demonstrating the scalability benefits of a novel spatial-awareness heuristic for fast server selection in dense environments. 3) ASTRA, an intelligent prefetching system for mobile AR that leverages object associations and user proximity, to achieve high cache hit rates. 4) Viveka, an activity-aware framework for wearables that co-optimizes sensor selection and sampling rates, yielding high energy savings with minimal accuracy impact. Collectively, these advancements provide a holistic approach to building efficient, responsive, and context-aware data management systems for the heterogeneous edge.