A video encoding speed-up architecture for cloud gaming
Mehdi Semsarzadeh, Mahdi Hemmati, Abbas Javadtalab, Abdulsalam Yassine, Shervin Shirmohammadi · 2014
In cloud-based video gaming systems, game engines are hosted in the cloud, and rendered gaming scenes are streamed to players over the Internet. In such systems, the tasks of rendering graphics and video encoding impose huge computational complexity on cloud servers. Therefore, speeding up the encoding process to meet the stringent requirements of the game becomes a critical issue in cloud-based video gaming systems. In this paper, we analyze the feasibility of developing a mechanism to accelerate the power-intensive process of video encoding, by using available game objects information in game engines. Specifically, we utilize the game engine's information about the motion of the objects within the scene in order to bypass the time-consuming procedure of Motion Estimation (ME) in conventional video encoders like H.264/AVC. Based on our analysis, the game engine's information could be usable inside a video encoder if an interface is involved to re-shape object information and make them compatible for the video encoder. Our experiments show that our approach accelerates the motion estimation process by 14.32% on average for two specific games, when object's information is taken into account during the encoding phase.