Research On Processor Energy Efficiency Optimization For Mobile Games

At present,the research on energy efficiency optimization of mobile processors lags behind the development of field technology.The mainstream manufacturers(such as Samsung,Huawei,etc.)still use the native DVFS(Dynamic Voltage Frequency Scaling)strategy of Linux kernel to achieve energy efficiency optimization by scheduling processor frequency on demand.However,there are some shortcomings,such as coarse frequency adjustment granularity(only supporting discrete frequency values),inability to recognize game scene changes,and inability to respond to game performance burst(burst computing performance improvement),etc.In this paper,aiming at the shortcomings of DVFS algorithm,we research on the optimization of processor energy efficiency during the mobile game runtime.First of all,aiming at the problem of coarse frequency adjustment granularity and inability to recognize scene changes,this paper proposes a hiGame-Governor frequency adjustment model based on M5 P model tree,which can accurately identify game scenes according to the software and hardware environment state(CPU frequency,load,memory reading and writing information,system temperature,voltage changes and game FPS,etc.)during the mobile game runtime.Through scene division,the regression equations between the frequency change rate and the system state variables are learned for different scenarios to realize the continuous frequency adjustment under the stable load change of mobile games.Secondly,in view of the problem that DVFS algorithm based on historical load prediction and frequency adjustment can't make accurate response to game performance burst,this paper proposes a frequency adjustment model Burst Cap for performance burst scenarios.By dividing the game performance burst scenario according to its corresponding system state,the corresponding burst subscene is learned through the heuristic search to learn its corresponding best burst frequency.Finally,a burst frequency lookup table BFT(Burst Frequency Table)is constructed to increase the frequency in time for the game performance burst to avoid frame loss and damaging the user experience due to insufficient performance.Finally,this article integrates the hiGame-Governor model and Burst Cap into the mobile game processor energy efficiency optimization model i Game Smooth.Through the Huawei Hi Silicon data acquisition software Hi Zee,a large number of real data of different mobile games are collected on the kirin 980 So Cs smartphone.Training and testing.And by comparing experiments with different algorithms,the energy efficiency optimization effect of the model proposed in this paper is verified.