Power Optimization Of Game Applications For Mobile Platform

Currently,low power has become an important topic in computer architecture,es-pecially on the mobile platforms,low power has become the bottleneck of the develop-ment of mobile platforms.Game applications,accounted for more than 50%of applica-tion downloads,are more hungry for power consumption due to its interaction-intensive characterization or computation-intensive characterization or graphic-intensive charac-terization.To effectively reduce the power of mobile platform during game playing,many researchers propose all kinds of power optimization strategies.However,there are less works covering the interaction characterization.In this paper,by analyzing the interac-tion nature and other natures of moible games,we introduce the game state information to direct the power optimizaiton for different components of mobile platforms.At the same time,to integrate user interaction,we propose and implement a game-specific benchamrk by adding the user interaction quantification.Our main work includes:1.According to the difference of user interaction and difference of user experience requirement in different game stages,we define game states.At the same time,to de-tect the runtime game states for close-source games,we propose a feature-based and a classifier-based game state detecting strategies,comparing their accuracy and overhead.2.We propose a automatic frame rate-based CPU power management policy.Since different game states require different user experience(i.e.,FPS in CPU)value,our scheme automatically detects the objective frame rate for every game state,then scales the CPU frequency using the objective FPS information to achieve high energy-efficiency.3.We introduce the game-state-based dynamic backlight dimming policy.The default policy during the game play adopts a static backlight.However,different game states require different user experience value.Our policy evaluates how backlight dim-ming influences the user experience,and how backlight dimming influences the display power.Then it scales the backlight according to the efficiency trade-off between the display power and QoS.4.We propose the PUBench which integrates pseudo user-interaction.By ap-plying our proposed Factors Profiling Architecture(FPA),PUBench uses the queuing model to imitate user interaction,and quantifies the APP-OS interaction and CPU-GPU interaction.The experimental results show that PUBench accurately covers the interac-tion characterization,workload characterization,and power characterization of games,eliminating the "user dependency".Finally,we give a case study of PUBench.