Optimizing Multi-grid Based Reduction for Efficient Scientific Data Management

Data management becomes increasingly important in dealing with the large amounts of data produced by today's large-scale scientific simulations and experimental instrumentation. Multi-grid compression algorithms provide a promising way to manage scientific data at scale, but are not tailored for performance and reduction quality. In this paper, we optimize a multi-grid based algorithm to achieve high-performance and high-quality error-controlled lossy compression. Our contributions are three-fold. 1) We quantize the multi-grid coefficients in a level-wise fashion, and leverage multi-grid decomposition as a preconditioner instead of standalone compressors to improve compression ratios. 2) We optimize the performance of multi-grid decomposition/recomposition with a series of techniques from both system-level and algorithm-level. 3) We evaluate our proposed method using four real-world scientific datasets and compare with several state-of-the-art lossy compressors. Experiments demonstrate that our optimizations improve the decomposition/recomposition performance of the existing multi-grid based approach by up to 70X, and the proposed compression method can improve compression ratio by up to 2X compared with the second best error-controlled lossy compressors under the same distortion.