Abstract: The complex structure and strong heterogeneity of the shielding model pose a challenge to the meshing and computational efficiency. Compared with standard Cartesian grid with the same modeling precision, non-uniform discontinuous grid transport sweeping algorithm can save computing resources, but the single grid computing cost is high due to the complex sweeping algorithm. The parallelization of non-uniform discontinuous grid transport sweeping algorithm based on ARES by analyzing the data structure and generation process of discontinuous grid was realized in this paper. The spatial-angle region decomposition was carried to make parallel multitask computing come true. The overload factors were introduced to reduce heterogeneity of spatial grid caused by local refinement to some extent. The recursive logical search and sweeping algorithm based on octree grid structure was transformed into pipelining by preprocessing to ensure the convergence consistency between serial and parallel algorithms. Numerical results of VENUS-3 benchmark problem demonstrate that the results calculated by serial and parallel algorithm are identical, and the relative deviation between the calculated results and the measured data satisfies the requirement of shielding calculation. The parallel test results show that the increases of overload factor and angle aggregation factor are beneficial to improve the parallel efficiency.