NERSC “Visualization Greenbook” Future Visualization Needs of the Doe Computational Science Community Hosted at NERSC

Abstract

In this paper we present the findings and recommendations that emerged from a one-day workshop held at Lawrence Berkeley National Laboratory (LBNL) on June 5, 2002, in conjunction with the National Energy Research Scientific Computing (NERSC) User Group (NUG) Meeting. The motivation for this workshop was to solicit direct input from the application science community on the subject of visualization. The workshop speakers and participants included computational scientists from a cross-section of disciplines that use the NERSC facility, as well as visualization researchers from across the country. We asked the workshop contributors how they currently visualize their results, and how they would like to do visualization in the future. We were especially interested in each individual's view of how visualization tools and services could be improved in order to better meet the needs of future computational science projects. The outcome of this workshop is a set of findings and recommendations that are presented in more detail later in this paper, and are briefly summarized here. Scientific visualization is a crucial technological capability that plays an important role in understanding data created by computational science projects as well as experiments. In order to be effective, visualization technology should be easy to use for a non-expert. The term “easy to use” encompasses a number of different categories, including a short learning curve, tight integration with computational frameworks, availability on the desktop as well as the fixed visualization facility, tools that are tailored for each specific application domain, and low cost. Current visualization tools fall short in several key areas of capability. Few visualization tools are capable of processing large datasets, such as those commonly generated at NERSC. Better support for parallel visualization tools may prove useful in leveraging large parallel machines as visualization resources. Multivariate visualization - multiple grids, many species, and many dimensions - is needed in order to quickly gain insight into large datasets. Related “drill-down” capabilities, such as the ability to quickly move from macro to micro views (used in “data mining”), would be extremely helpful in understanding data but are missing from most visualization tools. Many application scientists perceive a conundrum when it comes to visualization support. Support for visualization within each individual program level is often inadequate or nonexistent due to funding constraints, yet support for visualization at the institutional level is also often inadequate or nonexistent. Better solutions are needed for remote visualization. Current approaches are further constrained by network bandwidth and access to resources. The proliferation of visualization tools and data formats poses challenges. Researchers must often master many different tools in order to achieve the desired results. Data format conversion is often required when moving between tools. Common data formats and frameworks for visualization tools are needed to reduce duplication of effort and better promote sharing of resources and results. Better communication is needed between the visualization and computational science communities. The computational scientists are often unaware of current trends and practices in the visualization community. By being more aware of the needs of the computational science community, the visualization research programs can be crafted so as to be more responsive to their needs. As a result of the workshop, we have developed a set of recommendations that can be summarized as follows: • Establish a coherent program that focuses on remote visualization. A remote visualization program should provide tools and infrastructure that can be used by multiple “virtual teams”. • Establish mechanisms whereby generally-applicable visualization technology is developed and deployed in a centralized fashion. • Develop a research program in interactive visualization with running codes that stresses the integrated design and development of coupled simulation-visualization methods. • Establish a research program in the areas of multi-field visualization and multi-dimensional data visualization. • Establish a research program in the area of automated data exploration for next generation petascale datasets.