Exploring the Spectrum of Dynamic Scheduling Algorithms for Scalable Distributed-MemoryRay Tracing.

Area of Science:

• Computer Graphics
• Parallel Computing
• Algorithm Design

Background:

• Existing ray tracing algorithms often rely on static scheduling, limiting their ability to handle datasets exceeding system memory.
• Efficient scheduling is crucial for in-situ rendering on large distributed memory parallel computers.

Purpose of the Study:

• To extend and evaluate a family of dynamic ray scheduling algorithms for in-situ rendering.
• To address the challenge of rendering datasets larger than aggregate system memory.
• To compare dynamic scheduling against traditional static scheduling schemes.

Main Methods:

• Developed and implemented a family of dynamic ray scheduling algorithms.
• Algorithms consider both ray state and data access patterns during computation scheduling.
• Compared three dynamic algorithms against two static scheduling schemes on large distributed memory parallel computers.

Main Results:

• The dynamic scheduling approach successfully rendered datasets larger than aggregate system memory, which was not possible with static methods.
• For problems fitting within aggregate memory but exceeding typical shared memory, the dynamic approach showed competitive performance.
• Demonstrated the feasibility and advantages of in-situ dynamic scheduling for large-scale ray tracing.

Conclusions:

• Dynamic ray scheduling algorithms offer a viable solution for rendering extremely large datasets on parallel systems.
• This approach overcomes limitations of static scheduling, enabling out-of-core rendering.
• The proposed dynamic methods are efficient and competitive for large-scale computer graphics applications.