An Extended Data-Flow Architecture for Data Analysis and Visualization

NOTE: ALL FIGURES NOT INCLUDED IN PS FILE -- FOR COMPLETE COPY CONTACT LLOYD TREINISH, lloydt@watson.ibm.com ---- Over the last several years a number of software systems that provide visual programming, which embodied a notion of data flow, have been implemented. They were created under the premise that this paradigm was simple enough for users that are not experienced programmers to build applications. It was further assumed that this approach would greatly simplify the implementation and prototyping of computer graphics, data analysis and visualization systems that are composed of varied and often complex tasks. However, as the visualization community matured and the users of these tools grew in their sophistication, efforts to apply these systems to problems of realistic size and complexity illustrated a number of deficiencies within the typical implementations. The challenge from the perspective of developing tools for data analysis and visualization based upon the data-flow paradigm is to preserve the virtues of such an approach while trying to minimize the inherent limitations embodied by the use of a naive data-flow execution model for the visual programs. An implementation, IBM Visualization Data Explorer, that enhances the idea of data flow to include capabilities necessary to support realistic problems, while continuing to supporting its traditional advantages is discussed. These extensions include graph evaluation, conditional execution and caching.