CISE Research Resources: Matching Advanced Visualization and Intelligent Data Mining to High-Performance Experimental Networks

University Of Illinois At Chicago, Chicago IL

Investigators

Abstract

EIA 02-24306 DeFanti, Thomas A. Grossman, Robert L.; Leigh, Jason; Nelson, Peter C.; Yu, Oliver University of Illinois Chicago CISE RR: Matching Advanced Visualization and Intelligent Data Mining to High-Performance Experimental Networks This project, developing techniques for advanced grid computing with focus on visualization and data-mining applications, aims at setting up a high-performance, high-bandwidth Grid matched with Visualization and Data Mining Research. The Grid is made up of Lambdas (clusters of PCs) that are connected by high-bandwidths connections to other clusters. Software and toolkits to enable high-performance applications will be built with the goal of attaining a high-speed optical metropolitan-area network to be used in the data mining and visualization applications. The project expands grid technology to infrastructure, protocols, network memory and distributed control, and applications in constraint logic programming. Since optical networking technology is rapidly migrating from ultra-expensive long-haul carrier implementations to affordable regional- and metro-area community networks, this project explores inherent flexibilities in these new technologies to support large-scale data, visualization, and collaboration-intensive applications with very advanced real-time demands. Thus, application-aware software and middleware will be created to help interconnect tomorrow's terascale-class machines with distributed petabyte data stores, remote sensors, instrumentation and visualization over gigabit/sec to terabit/sec networks. LambdaNodes, defined as PC clusters with storage and visualization coupled to like clusters by numerous wavelengths (called lambdas), will be used connected by lambda networks to create a prototype metropolitan-scale LambdaGrid. With the goal of optimally matching data mining and visualization to high-performance optical networks with e-Science and homeland security model applications as expected drivers (achieving a 10x or greater end-to-end improvement over today), the equipment and support enables the following: 1. Matching data-mining and visualization capabilities on clusters to emerging wavelength-rich networks, 2. Distributing parallel computation and rendering for high-resolution volume visualization, 3. Providing applications signaling and control of both electronically and optically switched lambdas, 4. Measuring and monitoring multi-gigabit circuits over multiple wavelengths, 5. Providing users with networks that have known and knowable bandwidth and latency, 6. Investigating high-availability/uninterrupted cluster computing to support time-critical collaborations, 7. Addressing real-time applications in security domain, and, eventually, security of the data as well, 8. Integrating metropolitan-scale LambdaGrids with the emerging Global Grid and the National TeraGrid, and 9. Incorporating distributed data mining and visualization into undergraduate African-American-centered coursework and research through the Virtual Harlem Project.