Interframe Compression of Water Dominant Molecular Dynamics Simulations

Molecular dynamics simulations are files containing atomic and molecular data elapsed over time. State-of-the-art in parallel computers are able to generate simulations containing up to 10 million atoms and a large number of frames. These simulations typically hold uncompressed atom coordinates stored in binary format. This report’s main contribution is the development of a system through which compression can be performed. Many simulations contain a large quantity of water molecules which some of our compression schemes attempt to exploit. The compressors that are covered in this report use the temporal properties of the simulation in order to compress files. These are often called Interframe compressors. Five Interframe compression schemes are implemented and tested on several datasets at three different quantisations: 8-bit, 12-bit and 16-bit levels. The datasets were chosen based on properties such as size and the coherence of the motion in the simulation. The test results show that simple delta-based encoding performs well at all quantisation levels. At 8-bit quantisation delta encoding is able to achieve an average of approximately 10% compression rate on across all datasets. However, in the coherent datasets, a much greater compression rate is achieved. A trend that was picked up on was that the greater the number of previous frames, the worse the Interframe compression schemes tend to perform.