Distributed high-fidelity graphics using P2P

Abstract

The goal of physically-based rendering is that of obtaining very high visual fidelity for applications as diverse as defence and entertainment. Physically-based rendering accurately simulates lighting in virtual environments, and this simulation is computationally expensive. One common way to speed up rendering is by means of parallel computation, which is currently achievable on most desktop PCs due to multicore architectures. Computing accurate renderings in reasonable time, however, requires the use of the distributed memory model - historically using a client/server approach - and this kind of setup is not available to the general public. One alternative to the client/server approach is the concept of peer-to-peer (P2P) computing which allows loosely coupled peers sharing a common goal to share resources between them. While P2P has been used, for example, to stream geometry, its use in the context of high-fidelity rendering has been virtually absent. This work presents methods for sharing computed aspects of the illumination amongst peers that are visiting or have already visited the same virtual environment. By decoupling aspects of the lighting that are independent from the view point of any given peer, such information can be distributed to other peers that may not have computed it yet. The P2P solution allows faster rendering of high fidelity graphics in shared environments to obtain performances unobtainable by a single system.