| 摘要: |
| 立即模式图形并行绘制系统多采用sort-first算法，每绘制一帧都要求重发所有图元绘制指令，容易造成网络堵塞．保留模式并行绘制系统多要求集群机每个节点保留一份完整的数据拷贝，能改善网络堵塞，它可采用sort-first和sort-last并行绘制算法．保留模式并行绘制系统每个节点保留一份完整数据拷贝，使模型规模受单个节点内存容量限制．本文设计的称作In-the-Core的系统是混合sort-first和sort-last型保留模式并行绘制系统，将模型数据剖分并分布到绘制节点以充分利用集群机内存总量，用时间反馈的视点相关重分布算法保持负载平衡，有效减少了网络通信量.经验证此系统在绘制超大数据量模型时表现优秀． |
| 关键词: 并行绘制 In-the-Core 基本粒子 时间反馈 视点相关重分布 |
| DOI： |
| 分类号: |
| 基金项目:Supported by the National Grand Fundamental Research 973 Program of China under Grant No.2002CB312105(国家重点基础研究发展规划(973)) |
|
| An In-the-Core Parallel Graphics Rendering System for Extreme Large Data Sets Based on Retained-Mode |
|
PENG Hao-Yu,JIN Zhe-Fan,SHI Jiao-Ying
|
| Abstract: |
| We present a hybrid sort-first and sort-last parallel rendering system based on．retained-mode，called In-the-Core．which can render extreme large modeis on PC clusters．Traditional parallel systems have to issue all primitive-instructions when draw a new flame，or to copy whole scene on each node of cluster in advance．The approach can utilize full main memory capacity of cluster and lower the traffic of network．The main method is to partition a large model into smaller structures of elementary particle and distribmes them into the main memory of rendering nodes；meanwhile the load-balance is kept by view-depend partition algorithm with overhead feedback．Experiments on a cluster of 8 nodes show that the In-the-Core system can render models about 30～40M triangles at interactive frame-rate. It shows that a cluster of inexpensive PCs is all attractive alternative to those high-end graphics workstations. |
| Key words: parallel rendering In-the-Core elementary particle overhead feedback view-depend redistribution |
