Sudoku: Scalable High-Density Cloud Rendering Multi-Client Architecture

Nowadays, WebRTC, a transmitted protocol key for real-time interactions, is widely adopted in cloud-rendering areas, especially for the video-streaming part. Traditional end-to-end cloud rendering clients are becoming inadequate due to increasing server densities, particularly in areas like offline eSports and commercial promotions. Addressing this, we introduce Sudoku, a novel WebRTC-based solution for high-density cloud rendering. This implementation stands out by segregating the control and data planes. The control plane, on the CPU, adopts a single-process strategy, eliminating inter-process communication and enhancing scalability. In contrast, the data plane on the GPU introduces a fine-grained dataflow control, isolating each game stream to minimize resource contention and improve stream playback quality. Sudoku also integrates direct rendering technology, leveraging data locality for faster decoding and rendering. Our evaluations reveal that Sudoku dramatically surpasses existing models, cutting decoding and rendering times by 93.3% and 52.6%, respectively, and achieving a 5.76-fold density increase. So far, Sudoku has already been used at Intel for cloud-rendering performance tests and demonstrates its efficacy and potential for broader applications. This development signifies a major leap in the field, offering improved efficiency and scalability for high-density cloud rendering setups.