G-Match: A Fast GPU-Friendly Data Compression Algorithm

Data compression plays an important role in the era of big data; however, such compression is typically one of the bottlenecks of a massive data processing system due to intensive computing and memory access. In this paper, we propose a high-speed GPU-friendly data compression algorithm called G-mat...

Full description

Saved in:
Bibliographic Details
Published in:2019 IEEE 21st International Conference on High Performance Computing and Communications; IEEE 17th International Conference on Smart City; IEEE 5th International Conference on Data Science and Systems (HPCC/SmartCity/DSS) pp. 788 - 795
Main Authors: Lu, Li, Hua, Bei
Format: Conference Proceeding
Language:English
Published: IEEE 01.08.2019
Subjects:
Algorithm Parallelization
Compression algorithms
Data compression
Data Compression Algorithm
Dictionaries
Graphics Processing Unit(GPU)
Graphics processing units
Instruction sets
Parallel processing
Task analysis
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Data compression plays an important role in the era of big data; however, such compression is typically one of the bottlenecks of a massive data processing system due to intensive computing and memory access. In this paper, we propose a high-speed GPU-friendly data compression algorithm called G-match that takes full advantage of the GPU parallel computing power to speed up the compression process. The greatest challenge here is to solve the contradiction between the high data dependency inherent in the compression algorithm and the GPU single-instruction multiple-thread operating model. G-match achieves a high parallel degree by eliminating fine-grained data dependency and all path divergences in the algorithm. Compared with other, similar work on GPUs, G-match is the first thoroughly parallelized data compression algorithm. Experiments comparing other GPU compression algorithms show that G-match achieves approximately 33% speedup over the current fastest implementation and the highest compression ratio.
DOI:10.1109/HPCC/SmartCity/DSS.2019.00116