Scientific data visualization to convert simulation or observational data into images is one effective technique to intuitively understand the scientific features and meanings included in such data. This paper reviews the current status of research into largescale data visualization, including time-varying visualization, remote visualization, visualization using leading edge environments and optimization of the processes in a visualization pipeline. Visualization of a massive dataset produced by a large-scale numerical simulation takes a great deal of time and is not easy to perform on a single computer. Typical approaches to efficiently visualize such datasets are the parallelization of each process in the visualization pipeline such as filtering, mapping and rendering, as well as the optimization of the data structure. In the visualization of a time-varying dataset, data I/O processing occurs at every timestep of the dataset. Several methods of pre-fetching, parallel I/O and parallel pipeline processing have been developed as efficient data I/O techniques for time-varying visualization. In particular, parallel pipeline processing is widely used as an effective method that can reduce I/O bottlenecks and realize lower inter-frame delay. Remote visualization over a network enables users to utilize available computation resources and obtain visualization results on a desktop computer from the data that is retrieved from servers on the local network or over the Internet. In particular, distributed visualization, which is one of the configurations of remote visualization, is proposed in order to handle multiple datasets at remote sites. Collaborative visualization, also a configuration of remote visualization, allows multiple collaborative users to take part in several levels of visualization process. Finally, various visualization methods using leading edge environments are described. The utilization of massively parallel supercomputers and multiple GPU systems with tightlycoupled interconnecting backbones and a massive number of cores is effective to facilitate the faster visualization of larger datasets. In addition, virtual reality systems that enable users to interactively analyze such large-scale visualization results are also presented.
