Ugur Cayoglu (Karlsruhe Institute of Technology (KIT)), Frank Tristram (Karlsruhe Institute of Technology (KIT)), Jörg Meyer (Karlsruhe Institute of Technology (KIT)), Jennifer Schröter (Karlsruhe Institute of Technology (KIT)), Tobias Kerzenmacher (Karlsruhe Institute of Technology (KIT)), Peter Braesicke (Karlsruhe Institute of Technology (KIT)), and Achim Streit (Karlsruhe Institute of Technology (KIT))
The increase in compute power and more sophisticated simulation models with higher resolution output triggered the need for compression algorithms for scientific data. Currently there are several compression algorithms under development for. Most of these algorithms are using prediction-based compression methods, where each value is predicted and the residual to the true value is saved on disk. There are two forms of residual calculation which are currently established: Exclusive-or and numerical difference. In this paper we will summarize both techniques and show their strengths and weaknesses. We will show that shifting the prediction and true value to a binary number with certain properties results into a better compression ratio with minimal additional computational costs. This gain in compression ratio enables the usage of less sophisticated prediction algorithms to achieve the same or better compression ratio with higher throughput during compression and decompression. Further we will introduce a new encoding scheme to achieve a 8% increase in compression ratio on average compared to state of the art.
