Proactive Device-Wide Resource Throttling to Prevent System-Level ANRs in Peak-Load Commerce Applications

Abstract

Application Not Responding (ANR) events remain one of the most critical failure modes in Android-based commerce applications, particularly under peak-load conditions such as flash sales, seasonal promotions, and high-concurrency checkout events. While modern Android frameworks provide tools for asynchronous execution and background scheduling, these mech-anisms primarily operate at the application or thread level and lack awareness of device-wide resource contention. As a result, even well-architected applications can trigger system-level ANRs when CPU, memory, I/O, and binder resources become saturated concurrently.

This paper introduces a Proactive Device-Wide Resource Throttling (PDWRT) framework designed to prevent system-level ANRs by dynamically regulating resource consumption across the entire application process before critical thresholds are reached. Unlike reactive watchdog-based approaches, PDWRT continuously observes runtime signals—including main-thread latency, binder queue depth, garbage collection pressure, and system scheduler load—and applies adaptive throttling strategies across foreground, transactional, and background workloads.

The proposed framework is implemented and evaluated in the context of large-scale Android commerce applications oper-ating under extreme peak-load scenarios. Experimental results demonstrate a significant reduction in ANR incidence, improved UI responsiveness, and increased system stability without com-promising user-perceived performance. The findings suggest that proactive, device-wide throttling represents a necessary evolution in mobile system resilience engineering.