Adaptive Receding Horizon Offloading Strategy Under Dynamic Environment

摘要

This letter studies the offloading decision among multiple devices in dynamic environment to augment the computation on a resource-constrained local device. The run-time environmental dynamics are modeled as random disturbances, where the proposed adaptive receding horizon offloading strategy (ARHOS) notices the performance deviation from profile data and adapts the discount factor and the decision window size based on the disturbance frequency. Then, given deterministic profile data in the decision window, we propose a multiobjective dynamic programming approach to minimize the estimating cost while satisfying latency requirements with best effort. Simulation result shows that, by adjusting window size and resolving offloading strategy when disturbance happens, ARHOS’s performance is significantly better than the static optimal offloading strategy.