Article Preview
TopIntroduction
The main benefits of cloud gaming (CG) include no need to download and install or click and play. CG technology also does not require expensive hardware configuration, and the game subscription model can reduce game costs. However, the price paid in exchange for these advantages is high bandwidth consumption because the essence of CG is a video transmission that emphasizes timely interaction.
According to Zhang et al. (2019), a single user’s recommended downstream bandwidth for an acceptable-quality gaming experience is 3 megabits per second (Mb/s). Take 1,080-pixel (P) resolution as an example: If the transmitted image is 24 bits deep, for a 30 frames per second (fps) game experience, the required bandwidth is 18.66 megabytes per second (MB/s) without compression. Limited by the 100MB bandwidth of the 4G network, it can guarantee only up to five clients for regular use simultaneously. In addition, if the gaming service provider uses a public cloud, such as Amazon Web Services (AWS), the charging standard for transferring data from the AWS EC2 server to all over the world is as large as $0.02 per GB (Amazon, 2022), so the uncompressed cloud game video stream will consume $1.26 per hour for data transfer only. Moreover, this number will quickly become unbearable when the number of users increases to hundreds of millions. On the other hand, heavy gaming traffic can cause congestion in the backbone network, seriously affecting the performance of other online services. Therefore, knowing how to compress the transmission bandwidth of cloud games is the key to improve the game experience and save infrastructure costs.
To tackle this challenge, we propose a novel architecture that uses AI-based compression and enhancement algorithms to minimize the transmission volume of game video without sacrificing the perceived experiences. The methodology behind the architecture is that by deploying a game-specific trained SR model to the edge side beforehand, the edge server essentially pre-saves game video enhancement information closer to the player, thereby allowing the edge server to send relative low-quality data and save a significant amount of bandwidth. The trade-off is that more computing resources on the player side are used for real-time enhancement, and this requirement can be perfectly satisfied by the edge computing paradigm in which computation is moved as close to the end users as possible. Here is a summary of the main contributions of this paper:
- •
We propose an innovative cloud-edge collaborative computing architecture for CG. This architecture fully uses the computing power advantages of the cloud and the edge. The computing power is exchanged to reduce the amount of transmitted data, achieving economic benefits.
- •
We describe how we implemented an end-to-end prototype gaming system FRSR. This system is the first one using state of the art FR and SR technology to evaluate the collaborative computing architecture. The SR model proposed in Wang et al. (2021) is customized to support multiple regions of interest (ROIs).
- •
We describe our thorough objective experiments on FRSR, which were compared to four other cloud gaming implementations for four different game genres and recorded metrics, including peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), bits per pixel (BPP), and processing time. The results demonstrate FRSR’s effectiveness on bandwidth reduction while maintaining roughly the same level of perceived quality.