This thesis introduces a buffer-based hybrid distributed rendering systems for cloud-native gaming applications. Traditional cloud gaming relies on remote data centers for full-frame rendering, which often leads to high network latency and operation costs. Moreover, these systems can become inaccessible during network disruptions. With the increasing computational power of modern end-user devices, it is possible to organize these devices into a micro-cloud within a local network to collaboratively handle rendering tasks. This approach maximizes the use of idle resources, reduces latency, and improves rendering quality.
Drawing inspiration from Ad-hoc Computing and utilizing the widely adopted deferred rendering technique in modern video game engines, this thesis presents a hybrid distributed rendering system that decentralizes the computation of intermediate rendering buffers, such as ambient occlusion, indirect illumination, and reflections, by offloading them to more capable devices within the local network. The computed buffers are subsequently streamed back to client device to augment the final rendering process.
A prototype of the proposed system was developed using the Unity Engine and its Render Streaming package. The system was evaluated through both objective quality metrics and subjective visual analysis. Experimental results demonstrate the feasibility, potential, and visual benefits of this distributed rendering strategy. This work offers a foundational framework for future exploration and development of fully decentralized, buffer-based distributed rendering systems for cloud-native gaming.
