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

Li Lu,Bei Hua
DOI: https://doi.org/10.1109/hpcc/smartcity/dss.2019.00116
2019-01-01
Abstract: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.
What problem does this paper attempt to address?