Ray tracing is an algorithm for representing 3D scenes realistically . Principle , simple and powerful, based on the laws of geometrical optics : the image observer sees is the result of interactions between photons 6mis by light sources and objects with geometric and optical properties. The use of a physics-based model allows a good approach the following natural phenomena : the illumination by the light sources , the light reflection on objects and the transmission of light through transparent objects . The algorithm derived from this model , however, has a major drawback : the processing time of an image is very important. Although many acceleration methods have been proposed to try to improve its execution speed , this problem remains very present. However, with the current development of distributed memory parallel computers and computer networks , a very promising way of acceleration grows : parallelization. In addition, the pooling of capabilities '' memory " of each component of a parallel architecture enables the treatment of scenes with a larger number of objects with properties more complex . Chapter 1 presents the algorithm ray tracing and sequential acceleration proposed in the literature . D appears that the most effective techniques are based either on the use of bounding volumes or on subdivisions of space. Despite these optimizations, parallelization remains the source acceleration with the greatest potential . chapter 2 identifies the different types of parallel architectures and exposes families of algorithms used for parallel ray tracing on parallel distributed memory machines . Two strategies enable distribution base data : 1. sending rays and sending objects In chapter 3, after presenting our constraints , we compare the parallel algorithms can respond through their modeling. The results will lead us to propose two new algorithms for parallelization based on a new type of flow . Finally, in Chapter 4 we present our experimental results and their analysis. Our tests are performed on massively parallel machines and on a network of workstations.
