QoE-aware Computation Offloading to Capture Energy-Latency-Pricing Tradeoff in Mobile Clouds

Abstract

Computation offloading in mobile clouds helps mobile users save energy and enhance performance via mobile-to-cloud migration of processing. Although there exist many approaches to computation offloading, they have not explicitly considered the energy-latency-pricing tradeoff from the viewpoint of mobile users' context, e.g., user tendency, the remaining battery level. This paper tries to capture the user-centric perspective via quality-of-experience (QoE), and formulates two important problems: mobile-to-cloud transmission scheduling of the offloaded task's data and offloading service class selection. Regarding transmission scheduling, we introduce a database-assisted optimal dynamic programming (DP) algorithm and then propose two suboptimal but computationally-efficient approximate DP algorithms, ADP and ADPe, based on the limited lookahead technique. Regarding service class selection, we consider multiple service classes with different computing power and service charge, and formulate an optimization problem to minimize the overall cost incurred during offloading. An extensive numerical analysis has revealed that ADP and ADPe achieve near-optimal performance incurring only 0.35% and 2.1% extra cost than the optimum on average, and enhances the QoE-aware cost by up to 2.38 times compared to the energy-only scheduling. In addition, our service class selection algorithm is shown to choose the best class according to user tendency and the remaining battery level.