Edge Computing for Computer Games by Offloading Physics Computation

Abstract

Realistic graphics and smooth experience in computer games come with the cost of increased computational requirements on the end-user devices. Emerging Cloud Gaming that enables executing the games on thin devices comes with its disadvantages such as susceptibility to network latency and the incurred cloud computing cost for the game service provider. The monolithic architecture of the game engines also presents an issue for cloud gaming where scaling efficiency in the cloud turns out to be limited. This paper proposes using edge computing principles to offload a subset of the local computations executed by games to a nearby edge server typically assigned for gaming applications. Specifically, we focus on physics computations since depending on the number of objects and their interactions modes this part may have considerable computational cost. In order to demonstrate the effectiveness of our approach we developed an edge gaming framework called Edge Physics Simulation (EPS) using the open-source game engine Bevy and the Rapier physics engine. We come up with an experiment setup in which a game scene with a high number of objects is executed using both standard local computation approach and using the proposed EPS method. In the experiments up to 8000 objects of varying shape complexities are employed to trigger significant computational load due to the collision detection process. Assessment metrics used are average physics computation time, resource consumption of local device and, the breakdown of the physics duration into its critical components such network time, simulation time and compression time. Our results show that EPS significantly reduces physics time compared to local execution. For the highest number of objects 75% reduction in physics computation time is reported where breakdown of physics time is further analyzed.