Research On High Performance Computing And Ultra Large Scale Pre-processing Of Four-dimensional Spring Model

The application of discrete numerical methods in rock mechanics is largely limited by its huge amount of calculation.With the development of computer,it is possible to solve this problem by means of high performance computing.Developing,testing,optimizing,applying and popularizing parallel numerical methods and software for high performance computing of rock mechanics is not only the need for further development of rock mechanics,but also the strategic plan for balanced development of high performance computing in China.In this paper,based on Open MP and CUDA,multi-core CPU parallel,GPU parallel and CPU-GPU hybrid parallel of 4D-LSM are realized.The basic analysis of 4D-LSM parallel performance is made from the number of threads,the type of problem solving,the computing scale and hardware resources.The influence of digital accuracy on 4D-LSM calculation results is studied;the bottleneck factor of 4D-LSM computation scale is found;and an Eccentric Fourdimensional Model(ECC4D)based on 4D-LSM and a fast method for large-scale preprocessing of 4D-LSM using finite element geometric model are proposed.The results show that 4D-LSM has a high degree of parallelism and can better solve rock engineering problems with high performance computing;GPU parallelism of 4D-LSM must use double precision;the limit calculation scale is limited by physical memory capacity,according to the current computing resources of Alibaba Cloud Computing,the calculation scale of 4D-LSM can reach one billion particles;Eccentric Fourdimensional Model(ECC4D)can characterize the non-linear rock materials,and breaks through the limitation of Poisson's ratio of the traditional lattice spring model is 0.25.The proposed pre-processing method can well construct large-scale 4D-LSM model on the premise of visual modeling,and solve the problem that the model size is limited by the apparent memory capacity.