COMBO: compressed block-wise out-of-core diffraction computation for tera-scale holography.

Area of Science:

• Optics and Photonics
• Computer Science
• High-Performance Computing

Background:

• Large-scale computer-generated holography (CGH) demands substantial memory, often exceeding system capacity.
• Out-of-core processing is a viable solution but suffers from I/O bottlenecks in diffraction computations.
• Efficiently handling large datasets is crucial for advancing CGH applications.

Purpose of the Study:

• To develop a novel out-of-core processing framework, COMBO, for accelerating large-scale diffraction computation in CGH.
• To overcome I/O limitations inherent in traditional out-of-core methods for CGH.
• To enable the generation of tera-scale holograms on memory-constrained systems.

Main Methods:

• Implemented a block-wise data handling strategy within the COMBO framework.
• Integrated GPU-accelerated compression techniques to improve I/O efficiency.
• Utilized multiple Solid State Drives (SSDs) to further optimize data transfer speeds.
• Designed COMBO as an out-of-core processing system for diffraction calculations.

Main Results:

• COMBO demonstrated up to 4.16 times performance improvement over existing out-of-core methods.
• High-quality holographic reconstructions were maintained throughout the process.
• Successfully generated a 256K hologram, requiring 4TB of computational space, using only 64GB of system memory.

Conclusions:

• The COMBO system effectively addresses memory and I/O bottlenecks in large-scale CGH.
• COMBO offers a practical and efficient solution for generating massive holograms on standard hardware.
• This advancement has significant implications for memory-intensive holographic display and simulation technologies.