Tree-Based Adaptive Data Reduction Techniques for Scientific Simulation Data

Kurzfassung

Today's scientific simulations, for example in the high-performance exascale sector, produce huge amounts of data. Due to limited I/O bandwidth and available storage space, there is the necessity to reduce scientific data. Although, error-bounded lossy compression has been proven to be an effective approach tackling the trade-off between accuracy and storage space, many applications require that exact values need to be retained, e.g. check-pointing of simulations for restarting purposes. We are constructing novel lossless data compression approaches based on existing techniques for scalar data fields resulting from tree-based adaptive mesh refinement (AMR) applications and generalize their applicability to arbitrary multi-dimensional arrays. The data structures allow for progressive decompression until the initial resolution is losslessly reconstructed permitting direct access to coarser representations of the data which is favorably for visualization and transmission. We compare the proposed compression approaches against state-of-the-art compression techniques on benchmark data sets and exemplary on data originating from AMR applications.