A Scalable Hybrid Scheme for Ray-Casting of Unstructured Volume Data.

Area of Science:

• Computer Graphics
• Scientific Visualization
• High-Performance Computing

Background:

• Parallel volume rendering is crucial for visualizing large scientific datasets.
• Classical object-order and image-order techniques have limitations with complex data and load balancing.
• Unstructured grids present challenges due to complex boundary interleaving and variable cell sizes.

Purpose of the Study:

• To develop a novel parallel volume rendering algorithm.
• To address limitations in handling unstructured grids and load imbalance.
• To create an adaptable algorithm that performs efficiently across various rendering scenarios.

Main Methods:

• Developed a hybrid algorithm combining object-order and image-order rendering techniques.
• Designed the algorithm to operate on both structured and unstructured grids.
• Implemented workload division ensuring bounded computation per processing phase.
• Conducted extensive experiments varying camera position, dataset size, transfer function, image size, and processor count.

Main Results:

• The hybrid algorithm demonstrated strong performance across all tested conditions.
• Experiments scaled to 8,192 processors with datasets exceeding one billion cells.
• The algorithm effectively managed complex block boundary interleaving.
• Load imbalance issues arising from data or camera transformations were handled effectively.

Conclusions:

• The hybrid algorithm offers a robust solution for parallel volume rendering.
• Its adaptability allows it to function as either object-order or image-order rendering when most efficient.
• The approach is suitable for large-scale scientific visualization tasks.
• The method shows significant promise for improving rendering efficiency and scalability.