中文摘要：

科学可视化是高性能计算中分析数据的关键方法．千万亿次超级计算机提供的强劲计算能力使得科学家可研究更复杂的模型、进行更大规模的模拟计算，千万亿次的计算将产生TB／PB乃至EB级超大规模复杂时序数据，但由于I／O和存储的限制，传统的可视化模式已无法应对如此庞大数据的可视化分析．原位可视化将计算与可视化处理紧密结合，计算结果不经存储而直接在模拟所在的计算节点原位可视化处理成图片或提取特征数据，从而大幅减少存储、传输和后续处理的数据量．原位可视化被认为是分析千万亿次计算数据的最有效途径．文中从原位的数据压缩、原位的特征提取与跟踪，以及原位的可视化绘制3个方面对原位可视化技术进行了总结与归纳，并指出需要进一步探索的研究方向．

英文摘要：

Scientific visualization is indispensable in high performance computing for data analysis. Computing capacity provided by Peta-scale supercomputer enables scientists to research much more complex model and to do much larger scaled simulation, which produce huge complex time-vary data at TB]PB scale or EB scale. Due to the limit of I/O width and storage capacity, traditional visualization mode is unable to process such huge data. In-situ visualization combines simulation and visualization computation together, by which the data is directly processed in the exact computer node by rendering in an image or feature extraction. So the data needed to transfer or store is largely reduced. In-situ visualization is possibly the most efficient way to deal with the data produced by Peta-scale computation. In this paper, the related work on in-situ visualization, including in-situ data organization and compression, in-situ feature extraction and tracing, in-situ rendering, is summarized. At last, future work of this research topic is also discussed.