Research Of Data Management For Complex Scene In Large Scale Rendering

Today, realistic rendering is getting more and more widely used in film, animation, advertising and other industries. With the computer software and hardware performance boost, we need more at effect and speed of realistic rendering. Recent years, with physically limitation of development of CPU, parallel computing progress rapidly, large-scale rendering is going on its time. For example, movie<Avatar> spends 72 hours for single frame,16000 years are needed for the whole period with single core. However, parallel-rendering cut down it, weta studio spend several months with 35000 computing core,4000 work-station.From storage of data, after mesh created, it cannot be rendered directly, because these meshes is organized by specific software, for example, some software take account operations of modeling workers. However, when we need to render these meshes, we cannot rely on these data. It affects its compatible. So, we need a unified rendering data, and we provide plugins for modeling software to export standard data for us. According to large scale parallel rendering, we get some problems in efficacy, include multiple frames with complex scene produce a large amount of data, mainly in geometry; second comes a big number of small files. Through the work of this paper, the normalized data organization method can reduce the amount of data output and reduce inside and outside storage scheduling in the parallel rendering. For complex scenes with a large number of geometry, and a geometry has a material, we will divide it into 2 part, one for geometry and another for shader.1.Problems in geometry data in scene:As mid-data describe all the information of geometry that include light source and meshes, so, it has something to do with complex of scene. For example, a scene takes 1GB with 1000 geometries each of which takes 1MB. Moreover, if we render an animation,30 frame a second, only 10 minutes takes 300GB. I dispatch data of geometry and coordinate transformation, and keep all the reduplications in a file, so mass of data is cut down.2.Materials:After cut down the output data of geometry, the way of loading data affect the efficacy at the same time. As many kinds of materials that include 300 shaders in a scene with 1000 geometry exist, moreover,every shader of materials seem to be small to tens of several KB to tens of KB, It produces many small files if it does not works well. Moreover,when it comes to large scale rendering, as frequently I/O,problems of parallel is going on and it will affect interrupt of logical units,then comes cut of efficacy of parallel. This paper provide away to reorganize data of presentation of material in disks, make sure be load in memory correctly; second, according to data of materials stay in a great many small files,once load in memory,when material is needed,it can be load from memory directly.Most of I/Os are cut down especially for cluster with separate cache.All in all, by this paper, standard data organization will cut down data output and I/O when it works on parallel rendering. Complex scene has many geometries, and every geometry has a material.We will discuss these by two part at following section.In terms of implementation, based on the existing system, we use standard renderman to represent data, at the same time, we use modeling software building complex scenes, and get all the data reasonably organized, finally output data as far as possible to the following requirements:amount of data is as little as possible, data read as fast as possible and reduce system I/O.