The article presents MGARD, a software that provides MultiGrid Adaptive Reduction for floating-point scientific data on structured and unstructured grids. It is optimized for high-performance, error-controlled data compression and refactoring. The software is designed to address challenges such as storage reduction, high-performance I/O, and in-situ data analysis. It operates across different computing architectures, with GPU-optimized kernels and efficient memory and device management mechanisms. The software is a response to large-scale scientific applications that generate data volumes surpassing network and storage system capabilities.
Publication date: 12 Jan 2024
Project Page: https://arxiv.org/abs/2401.05994
Paper: https://arxiv.org/pdf/2401.05994
