Exploring Level-Wise Interpolation to Improve Lossy Compression Ratio for AMR Applications

Yida Li, Huizhang Luo, Chubo Liu, Kenli Li

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Adaptive Mesh Refinement (AMR) is widely adopted in High-Performance Computing (HPC) systems. However, none of existing AMR storage solutions has considered the high similarities among the adjacent AMR levels, which leads to a lower storage efficiency. In this paper, we propose level-wise data interpolation techniques to further reduce the storage of AMR applications. In particular, it generates finer level data based on coarser levels. Then, the differences (deltas) between the interpolated data and original data are stored to achieve a higher compression ratio for lossy compressors. We firstly use median absolute deviation and standard deviation to decide which interpolations are adopted. After that, we evaluate the effectiveness of level-wise interpolation with ZFP and SZ lossy compressors. The experimental results show that the compression ratio of deltas are improved up to 3 × compared to directly compressing the finer level data.

Original languageEnglish (US)
Title of host publicationLecture Notes on Data Engineering and Communications Technologies
PublisherSpringer Science and Business Media Deutschland GmbH
Pages259-266
Number of pages8
DOIs
StatePublished - 2022
Externally publishedYes

Publication series

NameLecture Notes on Data Engineering and Communications Technologies
Volume89
ISSN (Print)2367-4512
ISSN (Electronic)2367-4520

All Science Journal Classification (ASJC) codes

  • Information Systems
  • Media Technology
  • Computer Science Applications
  • Computer Networks and Communications
  • Electrical and Electronic Engineering

Keywords

  • Adaptive mesh refinement
  • Data reduction
  • High-performance computing
  • Level-wise interpolation

Fingerprint

Dive into the research topics of 'Exploring Level-Wise Interpolation to Improve Lossy Compression Ratio for AMR Applications'. Together they form a unique fingerprint.

Cite this