Data-Aware Predictive Scheduling for Distributed-Memory Ray Tracing

Area of Science:

• Computer Graphics
• Scientific Visualization
• High-Performance Computing

Background:

• Current ray tracing methods struggle with efficiency in partitioned data due to fixed scheduling.
• Existing ray scheduling techniques lack adaptivity and synchronicity, limiting performance gains.

Purpose of the Study:

• To develop a novel, adaptive ray scheduling method for efficient ray tracing.
• To overcome limitations of synchronicity and non-adaptivity in prior rendering techniques.

Main Methods:

• Incorporation of prediction models for dynamic adjustment of speculation levels in ray-data queries.
• Organization of rays in a tree of speculation nodes with pairwise coordination within adaptive ray groups.
• Implementation on a distributed system for concurrency and parallelism.

Main Results:

• Achieved up to three times higher throughput for volume and geometry rendering compared to non-predictive methods.
• Demonstrated high adaptability to various scene characteristics through dynamic speculation.
• Facilitated concurrency and parallelism via coordinated ray grouping.

Conclusions:

• The proposed adaptive ray scheduling method significantly enhances rendering performance.
• The approach is suitable for both interactive and offline rendering applications.
• Relaxing constraints of synchronicity and non-adaptivity leads to substantial throughput improvements.