What is it about?
This paper introduces LibPressio-Predict a library that makes it easy to compare different compression ratio prediction schemes. The paper then uses the library to compare 3 existing state-of-the-art prediction schemes on multiple compressors to show the trade-offs (e.g. accuracy, speed) between them on a level playing field since different papers use different datasets and often do not include other prediction schemes making comparisons difficult.
Featured Image
Photo by Myriam Jessier on Unsplash
Why is it important?
This library provides some basic tooling to make research in this space easier by providing a standardized benchmark suite that runs scalably on HPC systems allowing researchers to compare approaches quickly. It is also easy to extend with new methods. More importantly, it provides a standardized way to use prediction methods which enables building other tools on top of prediction methods without requiring a full understanding of them and without code changes as new methods are developed.
Perspectives
Increasingly it is important to not just use lossy compression, but not know in advance how it will perform to accelerate applications. Having a way to predict compressor performance with a consistent interface makes it much easier to build software libraries and tools that use these predictions.
Robert Underwood
Argonne National Laboratory
Read the Original
This page is a summary of: LibPressio-Predict: Flexible and Fast Infrastructure For Inferring Compression Performance, November 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3624062.3625124.
You can read the full text:
Resources
Contributors
The following have contributed to this page
