Distributed Volume Rendering of Very Large Data on High-Resolution Scalable Displays

Schwarz, N.

Location: University of Illinois at Chicago, Chicago, IL
PDF: thesis_nicholasschwarz.pdf

Caption: Researcher analyzing a Purkinje neuron on the LambdaTable
Credit: Lance Long, EVL

This thesis presents a methodology for rendering very large volume data on scalable high-resolution displays using a distributed-memory cluster.

The methodology uses a multi-resolution octree, an image-order data distribution method, a distributed shared-memory data management system, a multi-level cache, and hardware accelerated rendering techniques to produce a solution that is scalable in terms of input size and output resolution.

An analytical cost model validated by experimental results predicts the system’s behavior. The methodology’s usefulness is demonstrated with a number of domain specific datasets.

The primary contributions of this thesis include:
<ol><li>A review of research in the field of volume rendering and parallel volume rendering.</li><li>A methodology for rendering very large volume data on scalable high-resolution displays using a commodity distributed-memory cluster of computers that scales with the size of input data and the output resolution.</li><li>An analytical model validated by experimental results that predicts the methodology’s behavior.</li><li>The application of this methodology to domain specific problems in the fields of bioscience, geoscience and medicine.</li></ol>

Citation: Schwarz, N., Distributed Volume Rendering of Very Large Data on High-Resolution Scalable Displays, Thesis in partial fulfillment of the requirement for the degree of Master of Science in Computer Science, University of Illinois at Chicago, Chicago, IL, 2007-10-29.