SC20 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Approachable Error Bounded Lossy Compression

Author: Robert R. Underwood (Clemson University)

Advisor: Amy Apon (Clemson University), Jon C. Calhoun (Clemson University)

Abstract: Compression is commonly used in HPC applications to move and store data. Traditional lossless compression, however, does not provide adequate compression of floating point data often found in scientific codes. Recently, researchers and scientists have turned to lossy compression techniques that approximate the original data rather than reproduce it in order to achieve desired levels of compression. Typical lossy compressors do not bound the errors introduced into the data, leading to the development of error bounded lossy compressors (EBLC). These tools provide the desired levels of compression as mathematical guarantees on the errors introduced. The current state of EBLC leaves much to be desired. The existing EBLC all have different interfaces requiring codes to be changed to adopt new techniques; EBLC have many more configuration options than their predecessors, making them more difficult to use; and EBLC typically bound quantities like point wise errors rather than higher level metrics such as spectra, p-values, or test statistics that scientists typically use. My dissertation aims to provide a uniform interface to compression and to develop tools to allow application scientists to understand and apply EBLC. This canvas presents three groups of work: LibPressio, a standard interface for compression and analysis; FRaZ/LibPressio-Opt frameworks for the automated configuration of compressors using LibPressio; and work on tools for analyzing errors in particular domains.

Thesis Canvas: pdf

Presentation: pdf

Back to Doctoral Showcase Archive Listing