Scalable Data Transformations for Low-Latency Large-Scale Data Analysis

Interactive analysis of simulation results has become a mainstay of science and engineering. With continually increasing compute power, the size of simulation results continues to grow. However, network and mass storage device throughput are not increasing as quickly. This introduces difficulties in scaling analysis workflows to take advantage of these new compute resources.This dissertation describes a body of work that increases the scalability of interactive volume analysis workflows by moving elements strongly dependent on the input data size from the interactive phase of the workflow to the data preparation phase. This reduces the overall computational complexity of the interactive phase, enabling reduced interaction latency. Two related groups of approaches are explored: salience-aware techniques, and techniques for scalable salience discovery.Salience-aware techniques leverage the tendency for different parts of volume data to be of differing importance. In this dissertation, salience-aware techniques are proposed for load-balancing of isosurfacing on clusters, and salience-aware level of detail selection. In many cases, the salience of data may not be known a priori. Salience discovery techniques seek to facilitate the discovery of salience of different interval volumes. In this dissertation, a technique for iterative salience discovery, in the context of interactive transfer function design on large-scale volumes, is discussed. Supporting that, a technique is described for evaluating distribution range queries.For both groups of techniques, scalable data transformations are described and target applications are explored. This work streamlines workflows for visualization of large-scale volume data.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

doctoral

Discipline thesis:degree_discipline

Computer Science and Engineering

Grantor dc:publisher

The Ohio State University

Year dc:date

2013

Author dc:creator

• Martin, Steven

Contributors dc:contributor

• Shen, Han-Wei

• Computer Science

Statement dc:rights

• unrestricted
• This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.

Language dc:language

English

Repository record dc:identifier

http://rave.ohiolink.edu/etdc/view?acc_num=osu1366108187

OAI identifier oai:identifier

oai:etd.ohiolink.edu:osu1366108187