Dr Anurag Gupta (Scientific Officer 'F') Mr Kislay Bhatt (Scientific Officer 'F')
The most fundamental task in the design and analysis of a nuclear reactor core is to find out the neutron distribution as a function of space, direction, energy and possibly time. The most accurate description of the average behavior of neutrons is given by the linear form of Boltzmann transport equation. Due to massive number of unknowns, the solution of the transport equation imposes severe demands on computer processors / memory and requires best of numerical and computational schemes. A code ATES3 (Anisotropic Transport Equation Solver in 3D) had been developed in BARC for the deterministic solution of 3-D steady-state neutron transport problems. The code makes use of advanced Krylov subspace based schemes for the solution. To use ATES3 for practical reactor core simulations, it has been parallelized on BARC’s ANUPAM Parallel Supercomputer using Message Passing Parallel Programming model. The most time consuming step in the ATES3 code is transport sweep, which was targeted for the parallelization. We used three different data-decomposition techniques for parallelization, namely, parallelization in angular variable, parallelization using Diagonal Sweep approach and in the third approach, we are applying optimization techniques in Diagonal Sweeping. In the paper we discuss the above parallelization techniques and present the speed-up and efficiency figures obtained with each approach.