Combined geometric data structure and parallel method matching the structure of high performance computer

Abstract

The invention discloses a combined geometric data structure and a parallel method matched with a high-performance computer structure. The combined geometric data structure consists of multiple geometric layers; each geometric layer is used for describing a whole particle geometric calculation region of a medium model; each geometric layer is decomposed into multiple geometric blocks in a same division manner; each geometric block is composed of multiple geometric units for describing the medium model; an algorithm used for performing particle transport calculation is packaged in the geometricblock through a tracker object; particles are stored and managed through a particle set; a corresponding relationship between a user object and the geometric block is established through a geometric block object; each MPI process corresponds to at least one geometric layer object; each OpenMP thread in the MPI process corresponds to one geometric layer object during Monte Carlo particle transportparallel calculation; and Monte Carlo particle transport multilevel mixed parallel calculation is realized. The Monte Carlo particle transport multilevel mixed parallel calculation is realized; and efficient parallel calculation of hundreds of thousands of kernels can be supported.

Description

Technical field [0001] The present invention relates to Monte Carlo particle transport technology, in particular to a combined geometric data structure and parallel method that matches the structure of a high-performance computer. Background technique [0002] In nuclear reactor design, critical safety analysis and shielding calculations, particle (neutron and photon) transport solutions are the core of numerical simulation calculations. The Monte Carlo method is one of the commonly used numerical simulation methods. First, it establishes a simulation model of the real medium (including geometric shapes and the composition of medium nuclides), and then simulates the movement of a large number of particles in the medium (including particle generation, migration, Collision with the nucleus of the medium and other physical processes) and make appropriate records, and finally perform statistical analysis on the records generated by all particles to obtain numerical results. [0003] B...

AI Technical Summary

Problems solved by technology

[0004] However, with the continuous deepening of numerical simulation and the requirement of refined physical modeling, the field of nuclear reactors has proposed fine numerical simulation of the whole core. The medium model contains tens of millions of geometric bodies, and each geometric body contains hundreds of nuclides. The memory requirement of the media model has reached more than tens of gigabytes, far exceeding the memory capacity of the current computer CPU core ratio. The memory capacity allocated by a process cannot replicate the entire media model, resulting in the inability to use the particle parallel Monte Carlo method to realize the calculation.

[0005] On the other hand, in order to achieve higher peak performance, the architecture of supercomputers is constantly changing and becoming more and more complex

Images