Interactive Cluster-Based Parallel Rendering System For Large Datasets

Visualization of large datasets is an important problem with a wide range of applications, including computer-aided design, virtual training, immersive virtual environment and so on. How to take advantage of the cluster to expand Large-Scale Scene Rendering from single machine to multi-machine, using parallel rendering for High-Resolution and Large-Size display is the key problem of this paper.This paper present two different applications, VMPDRender and TyrMMR, according to static scene and dynamic scene.First, we introduced the hardware system of the cluster, including the detailed hardware information of the cluster and the construction of the system. After that we introduced the interactive tool and present how to use them.In VMPDRender which is based on master-slave mode, the slave consists of all fuctions of Large-Scale Scene Rendering System, the master is responsible for the synchronization of all slave nodes in each frame. So in the realization of VMPDRender, the synchronization is the key problem. We first present the "Barrier Synchronization", then proposed an improved method for the synchronization problem in the parallel rendering. After that, we present the protocol for the synchronization part. In the end, we gave the result of the experiment which showed that VMPDRender has excellent extensibility, and its performance is nearly as well as the stand-alone version, that is, the cost of synchronization can be negligible.In TyrMMR which is based on the improved version of client-server mode, the client consists of most of functions, while the server is only responsible for the rendering and the management of local datasets.In each frame, the client determined the rendering set, then send the data to the server which needs this data. This kind of structure is designed for real-time simplification, so TyrMMR can support the rendering of dynamic scene. We present some main sub-module, including the serialization of the datasets, the network module designed for graphics system and the cache system in the server node. Finally, the experiments verified that the object in the large-scale scene can move in real-time. Through the test of the system performance, we proved that the system has considerable scalability.Though the multi-machine has lower performance than the stand-lone version, loss of performance is not obvious with the increase in the number of server nodes.