Game-Based Offloading Optimization for Streaming Tasks in Mobile Edge Computing

In artificial intelligence applications of image or speech recognition, data is generally collected to run the model inference on a streaming basis. Owning to the intensive computing needs for processing streaming tasks, it’s necessary to design an efficient offloading scheme for streaming task processing. In this paper, a scenario of three-tier Mobile Edge Computing (MEC) system is considered, where the streaming tasks arrive at mobile device could be either run locally or offloaded to the edge device with limited computing resource or to the cloud server. Within the considered scenario, with the offloading and processing of streaming tasks in the MEC is systematically modeled, a streaming task offloading problem in terms of task response time and energy consumption is investigated. With the game theoretical approach applied, the existence of mixed strategy Nash Equilibrium is proven. An iteration algorithm is proposed to optimize the mixed strategy profile until Nash Equilibrium is reached and compute the system average tasks overhead. The result of extensive experiments shows that the investigations can improve the utility for mobile devices and achieve superior computation offloading performance.