MPC: A Massively Parallel Compression Algorithm for Scientific Data

Abstract—Due to their high peak performance and energy effi-ciency, massively parallel accelerators such as GPUs are quickly spreading in high-performance computing, where large amounts of floating-point data are processed, transferred, and stored. Such environments can greatly benefit from data com-pression if done sufficiently quickly. Unfortunately, most con-ventional compression algorithms are unsuitable for highly parallel execution. In fact, it is generally unknown how to de-sign good compression algorithms for massively parallel sys-tems. To remedy this situation, we study 138,240 lossless com-pression algorithms for single- and double-precision floating-point values that are built exclusively from easily parallelizable components. We analyze the best of these algorithms, explain why they compress well, and derive the Massively Parallel Compression (MPC) algorithm from them. This novel algo-rithm requires almost no internal state, achieves heretofore un-reached compression ratios on several data sets, and roughly matches the best CPU-based algorithms in compression ratio while outperforming them by one to two orders of magnitude in throughput