cuSZ(+): Optimizing Error-Bounded Lossy Compression for Scientific Data on Modern GPUs

Student: Jiannan Tian (Washington State University)
Supervisor: Dingwen Tao (Washington State University)

Abstract: Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. With ever-emerging heterogeneous high-performance computing (HPC) architecture, GPU-accelerated error-bounded compressors (such as cuSZ and cuZFP) have been developed. However, they suffer from either low performance or low compression ratios. To this end, we propose cuSZ(+) to target both high compression ratios and throughputs. Furthermore, we identify that data smoothness is a vital factor for high compression throughputs. Our key contributions are fourfold: (1) We propose an efficient compression workflow to adaptively perform run-length encoding and/or variable-length encoding. (2) We derive Lorenzo reconstruction in decompression as multidimensional partial-sum computation and propose its well-formed GPU implementation. (3) We optimize essential kernels and scale to the state-of-the-art A100 GPU. (4) We evaluate cuSZ(+) using real-world HPC datasets on V100 and A100. Experiments show cuSZ(+) improves the compression throughputs and ratios by up to 18.4× and 5.3×, respectively, over cuSZ.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: PDF


Back to Poster Archive Listing