In situ lossy compression allowing user-controlled data loss cansignificantly reduce the I/O burden. For large-scale N-body simulations whereonly one snapshot can be compressed at a time, the lossy compression ratio isvery limited because of the fairly low spatial coherence of the particle data.In this work, we assess the state-of-the-art single-snapshot lossy compressiontechniques of two common N-body simulation models: cosmology and moleculardynamics. We design a series of novel optimization techniques based on the tworepresentative real-world N-body simulation codes. For molecular dynamicssimulation, we propose three compression modes (i.e., best speed, besttradeoff, best compression mode) that can refine the tradeoff between thecompression rate (a.k.a., speed/throughput) and ratio. For cosmologysimulation, we identify that our improved SZ is the best lossy compressor withrespect to both compression ratio and rate. Its compression ratio is higherthan the second-best compressor by 11% with comparable compression rate.Experiments with up to 1024 cores on the Blues supercomputer at Argonne showthat our proposed lossy compression method can reduce I/O time by 80% comparedwith writing data directly to a parallel file system and outperforms thesecond-best solution by 60%. Moreover, our proposed lossy compression methodshave the best rate-distortion with reasonable compression errors on the testedN-body simulation data compared with state-of-the-art compressors.
展开▼
