A High Speed Lossless Compression Algorithm Based on CPU and GPU Hybrid Platform

With the development of modern computer technology, the growth of purchase funds for storage equipment is far behind the growth of data. How to effectively use the limited storage resources to store data becomes the focus of our research and the data compression technology is the key to solve this problem. The traditional compression technology on the CPU platform could not meet the requirements for processing massive data in the compression rate and the energy cost. The GPU provides a new solution for its strong parallel computation ability. In this paper we present an implementation of Parallel Matching Lempel-Ziv-Storer-Szymanski (PMLZSS), a high speed lossless data compression algorithm by using CUDA framework. The basic idea of our implementation of the PMLZSS algorithm on GPUs is the introduction of a paralleled matrix matching. The data needed to be compressed are divided into multiple dictionary strings and pre-read strings as the vertical axis and horizontal axis of the matrices, respectively. All of the matrices are paralleled matched in the different blocks. Compared with the traditional serial CPU platform LZSS compression algorithm and BZIP2 compression algorithm, the experimental data shows that on the premise of the basic compression rate unchanged, relative to the serial LZSS, the compression speed of PMLZSS is improved about 16x, while to the BZIP2, about 12x.