Adaptive Remote Visualization System with Optimized Network Performance for Large Scale Scientific Data

• Main Author: Zhu, Mengxia
• Other Authors: S. Sitharama Iyengar
• Format: Others
• Language: en
• Published: LSU 2005
• Subjects: Computer Science
• Online Access: http://etd.lsu.edu/docs/available/etd-11142005-124444/

This dissertation discusses algorithmic and implementation aspects of an automatically configurable remote visualization system, which optimally decomposes and adaptively maps the visualization pipeline to a wide-area network. The first node typically serves as a data server that generates or stores raw data sets and a remote client resides on the last node equipped with a display device ranging from a personal desktop to a powerwall. Intermediate nodes can be located anywhere on the network and often include workstations, clusters, or custom rendering engines. We employ a regression model-based network daemon to estimate the effective bandwidth and minimal delay of a transport path using active traffic measurement. Data processing time is predicted for various visualization algorithms using block partition and statistical technique. Based on the link measurements, node characteristics, and module properties, we strategically organize visualization pipeline modules such as filtering, geometry generation, rendering, and display into groups, and dynamically assign them to appropriate network nodes to achieve minimal total delay for post-processing or maximal frame rate for streaming applications. We propose polynomial-time algorithms using the dynamic programming method to compute the optimal solutions for the problems of pipeline decomposition and network mapping under different constraints. A parallel based remote visualization system, which comprises a logical group of autonomous nodes that cooperate to enable sharing, selection, and aggregation of various types of resources distributed over a network, is implemented and deployed at geographically distributed nodes for experimental testing. Our system is capable of handling a complete spectrum of remote visualization tasks expertly including post processing, computational steering and wireless sensor network monitoring. Visualization functionalities such as isosurface, ray casting, streamline, linear integral convolution (LIC) are supported in our system. The proposed decomposition and mapping scheme is generic and can be applied to other network-oriented computation applications whose computing components form a linear arrangement.