Technical Paper Review: Are All Games Equally Cloud-Gaming-Friendly? An Electromyographic Approach

Transcription

1 Technical Paper Review: Are All Games Equally Cloud-Gaming-Friendly? An Electromyographic Approach Kumar Gaurav CS300 October 21, / 19

2 Overview 1 Introduction 2 Related Work 3 Approach 4 Results 5 Model 6 Applications 2 / 19

3 Introduction Introduction What is cloud gaming and how is it different from online gaming? Why cloud gaming? Does it depend on the type of game? Need to develop a model to predict game s real-time strictness 3 / 19

4 Introduction What is cloud gaming? Cloud gaming, also known as Games-on-Demand, is a technology which offloads the tasks of graphics rendering, in addition to computation and storage needs, into clouds. 4 / 19

5 Introduction Why cloud gaming? Cloud gaming technology makes any computer game playable on a thin client without previous worries of hardware requirements. Consequently, there are no software installation overheads or compatibility issues if players wish to try a game because all the hardware and software is provided in the data centers by the game operators. 5 / 19

6 Introduction Does it depend on the type of game? The investigation is based on the observation that some games seem more playable than others in the cloud despite the inevitable latency, which is caused by network delay and processing delays that occur at both the server and client.the players negative emotion against the latency is quantified based on the electrical potentials produced by their corrugator supercilii muscle. 6 / 19

7 Introduction Need to develop a model to predict game s real-time strictness With the proposed model, one can easily assess a cloud games sensibility to latency without conducting costly QoE measurement experiments. 7 / 19

8 Related Work Related Work In M. Jarschel, D. Schlosser, S. Scheuring, and T. Hossfeld, An evaluation of QoE in cloud gaming based on subjective tests, in Workshop on Future Internet and Next Generation Networks, Jun 2011, Jarschel conducted user studies to measure and model the QoE of OnLive during game play. They proposed a regression model to predict OnLives QoE according to network performance metrics and players experience and attitude towards the games. This work complements it by answering the following unexplored questions: Is the inevitable response latency due to cloud gaming architecture game-dependent in terms of user experience? If so, how can a games design affect its susceptibility to latency and how can we utilize such knowledge to improve the gaming experience and system resource allocation? 8 / 19

9 Approach Quantifying the Real-Time strictness Nine games of the three genres of ACT, FPS, and RPG were chosen. ACT games: LEGO Batman, Devil May Cry, and Sangoku Musou 57 FPS games: COD: World at War, F.E.A.R 2, and Unreal Tournament 3 RPG games: Ys Origin, Loki: Heroes of Mythology, and Torchlight 9 / 19

10 Approach QoE measurement Facial electromyography(femg) was used to measure player s experience. Thus, RS = Q/L where Q is the degraded QoE when certain degree of Latency L was introduced. 10 / 19

11 Approach Latency Emulation WH KEYBOARD LL and WH MOUSE LL hooks (by calling the SetWindowsHookEx API) in Windows was used to delay all keyboard and mouse inputs made by subjects. 11 / 19

12 Results Figure : The relationship between latency and average femg potentials. 12 / 19

13 Results Figure : The real-time strictness of the studied games. 13 / 19

14 Model Assuming that we have p subjects, each of which recorded a game play video containing f frames, and the jth frame of the video recorded by the ith subject have the motion vectors {mc ij1, mc ij2,..., mc ijm }, we compute the games screen dynamics as p f i=1 j=1 sd( mc ij1, mc ij2,..., mc ijm ) /inter-frame-time f p Here sd() is the standard deviation and the final term inter-frame-time is used to normalize the screen dynamics factor so that it is comparable across games with different frame rates. 14 / 19

15 Model Figure : A screen shot of Ys with its motion vectors overlaid. The motion vectors reveal how the macroblocks shifted since the last P- or I-frame in the recorded H.264 video. Because the avatar was moving downward, the motion vectors of the frame were generally pointing toward the top center on the screen. 15 / 19

16 Model The command heaviness is given by: command heaviness = screen dynamics/input rate A regression based predictive model is given by: realtime-time strictness = command heaviness 16 / 19

17 Model Figure : (left) The scatter plot of the actual and predicted RS of the 9 games. (right) The relative error of RS prediction in cross validation. 17 / 19

18 Applications Applications One can infer players having a worse gaming experience than others, and accordingly prioritize the servers resources, such as CPU and GPU, to reduce those players latencies and thereby mitigate QoE degradation they would otherwise experience One can estimate the total utility (by considering both operation cost and players satisfaction) and assign to each player a data center which provides a just good enough gaming experience while saving as much operation cost as possible. 18 / 19

19 Applications Thank you! 19 / 19