[Image] Visualization: User Needs DoD High-Performance Computing Modernization, PET Program Polly Baker PET Academic Lead, Visualization, WES/ASC/ARL NCSA, University of Illinois January 30, 1997 The first several months of the DoD HPCMO PET Program have presented opportunities to meet with a number of MSRC staff and users. This document identifies user needs in visualization and outlines areas where improvements would enhance the MSRC computing environment. 1 Visualization and High-Performance Computing The evolution of high-performance computing environments gives rise to new challenges for visualization. Users have expressed needs in the following areas: 1.1 Grid Generation New problems, and multi-disciplinary approaches to problems, give rise to issues of visual, interactive grid generation. 1.2 Postprocessing large data sets With bigger machines, users produce larger data sets. Commercial off-the-shelf visualization tools (AVS, NAG Explorer) generally cannot handle these very large data sets. 1.3 Real-time visualization Users want to be able to monitor simulation results as they are produced, requiring that visuals be generated concurrent with the simulation. This is useful in at least two contexts: 1. distributed compute and visualize. In many cases, it will be adequate to dump data sets from the simulation to some other machine that maps to graphical form for user analysis. This would be particularly useful for debugging, when the simulation runs are relatively small and perhaps running in an interactive queue. 2. co-processing environment for compute and visualize. In other cases, it is desirable to do the visualization step on the same machine as the computation. In this case, visualization algorithms may need to be parallelized, so that they can perform at a level that does not significantly slow the computation. 1.4 Interactive steering Application teams want to go beyond real-time visual monitoring and want the ability to interactively control parameters of a running simulation. 1.5 Building interactive interfaces Users want to be able to generate their own tools they can use in real-time visualization or interactive steering situations. Particularly aimed at debugging, these tools can employ relatively simple graphics and user interfaces. A supporting library for rapid development of simple visualization tools is needed. 1.6 Performance analysis In addition to analyzing simulation data, visualization is useful for analyzing simulation performance. As users try to improve performance of their codes, visual tools to aid performance analysis are needed. 2 Networked Visualization High-speed networking is an integral part of high-performance computing. 2.1 Remote users Off-site researchers should have easy access to visualization functionality - ideally, the same access enjoyed by on-site users. 2.2 Collaborative visualization HPC applications often involve geographically distributed teams. Tools to support collaborative visualization are needed, such as shared repositories of images and animations with multimedia annotations, etc. High-performance networks also make it possible to support synchronous visual exploration of data sets. Efforts are needed to define functionality needed for collaborative work and build implementations. 3 New Visualization Technologies 3.1 Immersive Technologies Users at ARL, ASC, and WES have all expressed interest in this emerging area. Ongoing efforts are underway at WES (BOOM) and ASC (molecular docking, CFD). Additionally, as part of PET, the 3 sites will soon have ImmersaDesks. Training in the use of this equipment is needed, as are specific ``pioneer'' projects that explore and demonstrate potential of the new technologies. Software development is also needed to: 1. build ImmersaDesk support for visualization. The library that accompanies the Desk deals with stereo rendering, reading the trackers, synchronizing processes. It has no constructs to support visualization (i.e, no code to generate and draw isosurfaces, vector plots, etc.) 2. provide cross-platform visualization. As much as possible, visualization applications that work on the ImmersaDesk should also operate on the desktop, providing interoperability across display environments. 3.2 Multi-modal interfaces Using auditory cues and/or haptic feedback to provide information about data has potential for some applications. Immersive applications in particular seem to benefit from multi-modal interfaces, but desktop tools can use these techniques also. Information gathering about products and potential applications is needed, along with ``pioneer'' projects. 3.3 Representation Issues Using alternative representations of a data set can reveal subtleties that go unnoticed when viewing a single representation. Demonstrations of what representations might be useful in an application area would expand users' horizons about how to explore their data visually. Development of new representations might be necessary in some cases. For application teams that are new to visual data analysis, side-by-side assistance with a visualization package is often needed. 4 Training The preceding discussion suggests courses in ImmersaDesk use, commercial packages, building interactive vis tools, performance analysis vis tools. Users have also expressed themselves regarding delivery method, as described here. 4.1 Application-specific use of visualization packages The general-purpose commerical visualization packages (AVS, NAG Explorer, IBM DX) are large and relatively complex. Users need some form of introductory training to their functionality and operation, and they would prefer that it be done specific to their discipline. 4.2 Just-in-time training The typical user can't invest a lot of time learning about visualization. Users want easy and timely access to just that information that will help them get the job done. This implies the need for a Web-based repository of visualization information and training material, complete with hands-on exercises. For visualization, this delivery method might prove more effective than on-site delivery of multi-day courses. ---------------------------------------------------------------------------- Polly Baker Sun May 18 19:22:18 CDT 1997