Authors & Affiliations

Goremykin S.A.
National Research Center “Kurchatov Institute”, Moscow, Russia

Abstract

Full-scale simulation of reactor dynamics with the coupled neutronic and thermo-hydraulic codes requires great computational efforts, most part of which usually refers to neutronic modeling. The given work is dedicated to super-computer parallelization of dynamic software package (SP) BARS. SP BARS makes it possible to calculate steady-state reactor conditions at any allowable power level, to simulate the reactor core cycle, the xenon transient and fast nominal and emergency transients with prompt and delayed neutrons. To speed up the computational process, parallel calculations are currently used. In this context we can distinguish two directions. One is good for “coarse-grained” tasks, in which a small number of parallel subtasks can be singled out, but each one is rather labor-consuming in terms of calculations. This is a trend of super-computers that combine a great number of full-scale processors within the framework of both a computer and a cluster). The second trend is suited for “fine-grained” tasks where a lot of small parallel subtasks can be distinguished. This is a trend for video cards (Nvidia CUDA) and pipelined coprocessors (Intel Xeon Phi). This paper considers the first approach. Different technologies are used for its implementation. In this case the OpenMP and MPI technologies will be considered.

Key words
Heterogeneous reactor, Fourier series, parallelization, monopole and dipole approximation, neutron flux, coarse-grained and fine-grained tasks

Article Text (PDF, in Russian)

References

UDC 621.039.5.021:004.94

Problems of Atomic Science and Technology. Series: Nuclear and Reactor Constants", issue 4:10, 2014