To perform the microscale modeling of granular materials the discrete element method (DEM) can be effectively applied. According to the DEM each particle is defined as a separate object. Usage of the DEM in the modeling of a large number of objects leads to the several challenges of the simulation results storage. The first problem is an extremely high volume of generated data. Each object has various time-dependent properties like coordinates, velocity, acceleration, angular velocity, etc. (Fig. 1).
All properties of objects are to be saved at any time and in case of simulation of a large number of objects, it definitely leads to the extremely high volume of generated data. Another problem is a loss of significant results. In order to reduce the size of stored data, the discretization is used. After the each specific time interval (saving time step), the whole state of the simulated process is to be saved. Usually, saving time step is to be defined by the user without previously analysis of the type of simulated process and it often leads to the loss of the important information about the objects dynamics. Fig. 2 illustrates the trajectories of a particle obtained after analysis of the simulation results with a different saving time steps using the linear interpolation. The increase of the saving time step leads to the loss of data about particle trajectory.
The main purpose is a development of advanced methods to store simulation results obtained after modeling of granular materials with discrete element method. The novel concept should allow to reduce the volume of saved data by several times and to avoid the loss of the important information. The developed methods should be implemented with C++ programming language as a subsystem for the data storage and integrated into the simulation framework MUSEN. General overview of the project objectives and planned work is schematically illustrated in Fig. 3
To reduce an amount of scientific datasets, depending on a type of the modeling process and behavior of objects the lossy and lossless methods of data compression can be applied. Lossless methods are used when the original data should be exactly reconstructed from the compressed data. Lossy methods include some loss of data but they are used when one needs to reach much higher compression rates for scientific data. For example: lossy methods can be effectively used for the visualization process, while for the restart simulation only lossless methods are to be used. Fig.4 shows the main strategy of applying of lossy and lossless methods for data saving. Additionally, for the compression of all saved data the binary file format as well as a protocol buffers serialization technology are to be applied. For the fast data access the reading and writing functions with a good performance and data buffering into the fast random-access memory (RAM) are used.
German Research Foundation (DFG)