A Simple and Real-Time Parallel Compression of Time Series Scientific Simulation Data for Interactive and Cooperative Supercomputing

In order to realize a cooperative supercomputing environment, Large-scale scientific simulation data should be exchanged among supercomputers over the Internet. To maintain the interactivity of the time-series simulation in such an environment, real-time data compression is required. Given the accuracy requirements of scientific simulation, we only focus on lossless data compression. In this paper, we propose Think Mantissa as Integer method to decrease the entropy of scientific datasets, achieving a good compression ratio and speeds. We evaluate this scheme in two experiments and compare it with the existing compression strategies (gzip, bzip2). Our approach gets about a 30% improvement in the compression ratio and hundreds of times speed acceleration. We also computed the total time of the data transfer using this method under some specific bandwidth cases, and nearly all of them have gotten a good acceleration.