Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

cuSZ-$i$: High-Ratio Scientific Lossy Compression on GPUs with Optimized Multi-Level Interpolation

Jinyang Liu, Jiannan Tian, Shixun Wu, Sheng Di, Boyuan Zhang, Robert Underwood, Yafan Huang, Jiajun Huang, Kai Zhao, Guanpeng Li, Dingwen Tao, Zizhong Chen, Franck Cappello

TL;DR

CUSZ-i is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module and is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module.

Abstract

Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. Compared to CPU-based compressors, GPU-based compressors exhibit substantially higher throughputs, fitting better for today's HPC applications. However, the critical limitations of existing GPU-based compressors are their low compression ratios and qualities, severely restricting their applicability. To overcome these, we introduce a new GPU-based error-bounded scientific lossy compressor named cuSZ-$i$, with the following contributions: (1) A novel GPU-optimized interpolation-based prediction method significantly improves the compression ratio and decompression data quality. (2) The Huffman encoding module in cuSZ-$i$ is optimized for better efficiency. (3) cuSZ-$i$ is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module. Evaluations show that cuSZ-$i$ significantly outperforms other latest GPU-based lossy compressors in compression ratio under the same error bound (hence, the desired quality), showcasing a 476% advantage over the second-best. This leads to cuSZ-$i$'s optimized performance in several real-world use cases.

cuSZ-$i$: High-Ratio Scientific Lossy Compression on GPUs with Optimized Multi-Level Interpolation

TL;DR

CUSZ-i is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module and is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module.

Abstract

Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. Compared to CPU-based compressors, GPU-based compressors exhibit substantially higher throughputs, fitting better for today's HPC applications. However, the critical limitations of existing GPU-based compressors are their low compression ratios and qualities, severely restricting their applicability. To overcome these, we introduce a new GPU-based error-bounded scientific lossy compressor named cuSZ-, with the following contributions: (1) A novel GPU-optimized interpolation-based prediction method significantly improves the compression ratio and decompression data quality. (2) The Huffman encoding module in cuSZ- is optimized for better efficiency. (3) cuSZ- is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module. Evaluations show that cuSZ- significantly outperforms other latest GPU-based lossy compressors in compression ratio under the same error bound (hence, the desired quality), showcasing a 476% advantage over the second-best. This leads to cuSZ-'s optimized performance in several real-world use cases.
Paper Structure (32 sections, 1 equation, 9 figures, 3 tables)

This paper contains 32 sections, 1 equation, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background and Research Motivation
  4. cuSZ Framework
  5. Research Motivations
  6. Limitation of existing GPU-based lossy compressors in data reconstruction quality
  7. Limitation of existing GPU-based lossy compressors due to lossless encoding modules
  8. Design goals of cuSZ-$i$
  9. cuSZ-$i$ Design Overview
  10. cuSZ-$i$ Interpolation-based Data Predictor
  11. Basic Design Concept of G-Interp
  12. Interpolation Configurations of G-Interp
  13. Interpolation splines
  14. Level-wise interpolation error bound
  15. Profiling-based Auto-tuning of G-Interp Interpolations
  16. ...and 17 more sections

Figures (9)

  • Figure 1: G-Interp, the parallelized interpolation-based data predictor.
  • Figure 2: Interpolation splines.
  • Figure 3: Illustration of spline interpolations using a 2D slice.
  • Figure 4: Showcase: counts of nonzero quant-code among CPU SZ3, GPU G-Interp, and GPU Lorenzo. Two relative-to-value-range error bounds are used on Miranda-Pressure. Dots in the $33\times9\times9$ bounding box indicate the nonzeros.
  • Figure 5: The PSNR advantage of interpolation over Lorenzo on two error bounds. One snapshot is selected for every 100 among 3700, excluding several ones corresponding to the simulation's initialization phase.
  • ...and 4 more figures