AWARE: Evaluating PriorityFresh Caching for Offline Emergency Warning Systems

TL;DR

The paper tackles the challenge of delivering the most actionable emergency alerts when connectivity is imperfect by proposing PriorityFresh, a semantic, actionability-first caching policy evaluated within the AWARE simulation framework. PriorityFresh uses a fixed-weight eviction score base(a,t) = $w_S s(a) + w_U u(a) + w_F f(a,t)$ with $f(a,t) = e^{-rac{1}{600}(t - a_{ ext{issued}})}$ and weights $w_S=4$, $w_U=5$, $w_F=5$ in experiments, aiming to surface urgent, fresh, and nearby guidance. The study finds that PriorityFresh improves actionability without reducing delivery efficiency across various cache sizes and network reliabilities, while other policies trade off freshness or timing stability. Context-aware guidance shows when to prefer PriorityFresh, PAFTinyLFU, or TTLOnly based on device capacity, network conditions, and operational objectives. Overall, the work advances offline-first emergency warning systems by demonstrating how content semantics and priority-aware caching can improve protective action readiness under real-world constraints.

Abstract

PriorityFresh is a semantic, actionability-first caching policy designed for offline emergency warning systems. Within the AWARE system's simulation environment, PriorityFresh optimizes which alerts to retain and surface under constrained connectivity. Experiments indicate improved actionability-first performance without harming efficiency. A separate Priority Forecasting model is used only to synthesize realistic alert sequences for controlled experiments and does not influence caching or push decisions.

Figures (11)

• Figure 1: Summary table of baseline metrics across policies.
• Figure 2: Cache-size sweep: winner heatmap across metrics/policies.
• Figure 3: Cache-size sweep: all-metrics grid overview.
• Figure 4: Network-reliability sweep: winner heatmap across metrics/policies.
• Figure 5: Network-reliability sweep: all-metrics grid overview.