Deadline-Aware Task Offloading under Intermittent Connectivity in Mobile Edge Networks

Abstract

Mobile edge applications increasingly depend on remote computation, but unstable wireless connectivity can make conventional latency-based offloading decisions unreliable. This paper presents a deadline-aware task offloading strategy designed for applications operating under intermittent network conditions. Instead of selecting an execution location solely from current bandwidth and server load, the proposed strategy estimates short-term connection reliability from recent transmission history and jointly considers task deadline, input size, computational demand, and queue state. We evaluate the approach using 4.7 million task requests generated from 312 mobile and IoT devices across 26 edge nodes, with network traces containing bandwidth fluctuations, handovers, and temporary disconnections. Under moderate network instability, the proposed method completes 93.1% of tasks within their deadlines, compared with 84.7% for latency-greedy offloading and 87.5% for a queue-aware baseline. During highly unstable periods, deadline violations are reduced by 31.8%, while average device energy consumption decreases by 14.6% relative to local-only execution. The scheduling decision requires 0.83 ms on average, allowing it to operate online without introducing substantial control overhead. Further analysis shows that connectivity prediction is most beneficial for medium-sized tasks whose transmission and execution costs are comparable, while very small and highly compute-intensive tasks generally favor local and edge execution, respectively.