Engineering and science computation is rapidly developing toward large-scale, high performance. Parallel computation is the future of high performance computation. In this paper we present a parallel computation model of three-dimensional rock failure process, and implement with a RFPA3D code developed from serial version. The part of visualization and analysis run on PC installed MS windows as the user interface platform. The code is implemented using VC and OpenGL. The most time-cosuming part-FEM compuation-is fully parallelized with MPI and runs on Linux cluster in order to make good use of multi-processor platrom compuation resource. DDM (domain decomposition method) is adopted as partitioner to partition a whole domain to server sub domains. And hostless structure is employed to use all computation resource. Parallel preconditioned iterative solver is utilized for solving the equation system. The part is coded with Fortran and C++. In order to make the two parts work together in coordination, we design client-server model. The one part on PC works as client and the other serve as server to supply computation resource. They are connected through network. We obtain the high parallel efficiency on the cluster system in our laboratory. The numerical example shows the code has very abroad applications in rock mechanics.
With supercomputer and supercomputing technology rapidly developing, high performance computation (HPC) becomes a trend and attracts great public attention. Many researchers have applied parallel computation to their numerical modeling because the need for computation of large-scale problems and high demand of accuracy in theory study and engineering analysis is continuously sharp increasing and multi-processor platforms provide an opportunity for faster performance than single processor platform. In particular in nonlinear structural analyses parallel computation has many applications(Watson 1996; Mark 2000; Gummadi 1997; Miyamura 2000; Nikishkov 1998; Romero 2000). But in the rock mechanics and engineering field, the use of parallel computers is still relatively new, especially in simulation of crack propagation. Dowding et al.(Dowding 1999) employed NURBM3DP, a three dimensional distinct element code, to calculate dynamic response of a cavern in jointed rock mass and compared performances of the code on different platforms. Washington et al.(Washington 2003) used an algorithm titled "TRUBAL for Parallel Machines(TPM)" based on discrete element method to simulate triaxial test and demonstrated the benefits and feasibility of adaptation of a massively parallel computer for micro-mechanical simulations. Chung and Kim(Chung 2003) developed a nonlinear parallel algorithm using sparse direct multifrontal solver. They performed analyses of damage localization problems for two and three dimensional crack models. Zhang et al.(Zhang 2004) used parallel finite element method (FEM) to analyze rock engineering problem on windows pc cluster. We know that understanding how fractures develop in materials, particularly in rocks, is a crucial problem to many disciplines such as material sciences, geological sciences and civil engineering. However, all the above mentioned studies were not involved parallel simulation of crack propagation, particularly for rocks which are highly heterogeneous.
