The Data Management Strategies Of The Large-scale Distributed Rendering

Rendering is a key process in the anime and films making, occupying 30% of t he creating time. Currently, in large-scale distributed rendering environment, rendering data is stored in remote disks. During the rendering, compute nodes need to read data from remote disks, in the large-scale rendering, the concurrent access of large a mounts of nodes will lead to the I/O bottleneck and data transmission congestion, making data reading speed undesirable and making the rendering time longger.In order to solve these problems, a distributed memory management an architecture called RenDMS is proposed. Free memories in computing nodes are used effectively and different data management strategies are designed in the architecture. The concept of rendering unit is proposed, so the large amount of nodes are divided into many units, one is master and others are slaves, which make up a RenDMS; the rendering unit make the concurrent access of the large amount of nodes to only the master access in each unit, decrease the concurrent access to the disk efficiently. And dynamic adaptive data placement strategy is used to allocate memories efficiently in different memory rat io situations. In different rendering scene, the ratio of lens data and the free memory is different, the memory used to store data is also different, dynamic adaptive data placement strategy take many factors into account, make the copies flexibly, dynamic adaptive data placement strategy contains fully copied strategy, fully distribute strategy and hot data strategy. After rendering calculat ion starts, the frame data prefetching strategy can overlap rendering computation and data transmission. According to the data prefetching strategy, we load the data which will be visited before the local memory, which will speed up the rendering.Experiments have shown that our work can speed up re ndering process and decrease the rendering time efficiently.