No abstract

The possibility of using geometrical approximations, maintaining local materials mass balance in each spatial grid cell by introducing additional mixtures for the cells where several initial materials are present, in kinematic calculations of neutron fields in a reactor core is analyzed. To prescribe a 3D geometry of the core, combinatorial geometry methods, implemented in the MCU computer program for obtaining a Monte Carlo solution of the transport equation, are used to convert the combinatorial formulation of the geometry into a grid representation -the ray tracing method. Calculations of a VVER-1000 core and a model of a spent fuel repository show that the method considered here gives a severalfold computational gain over standard approximations of the geometry.A direct kinetic calculation of neutron fields in a reactor is a topical problem, since it permits lifting the well-known limitations of the diffusion approximation and thereby increasing substantially the accuracy of calculations of the energy release, fuel burnup, and other characteristics which are of practical interest. Difficulties arise in such calculations because it is necessary to prescribe a three-dimensional geometry of the computational region, convert the geometry to a 3D difference grid with several million cells, and use efficient methods to accelerate internal iterations over the scattering integral in a group and external iterations over the region of neutron thermalization and the fission source when determining the effective neutron multiplication coefficient k eff . Another condition for obtaining the desired computational accuracy is the availability of a high-accuracy multigroup or problem-oriented system of constants, making it possible to take account of heterogeneity, to the desired degree of accuracy, when calculating the resonance blocking of cross-sections in fuel elements and other structural components of the core.As a result of the high heterogeneity of the core, attempts to use standard approximations of the geometry on a grid, including irregular grids, require for convergence dense three-dimensional grids with spacing 0.1 mm or less, which results in an unjustifiable increase in computational expenditures. The present article analyzes the possibility of using sparser three-dimensional grids by using approximations which maintain local materials mass balance (i.e., in each spatial cell of a grid covering the computational region) by introducing additional mixtures in cells with several initial materials. In this approach, known as the VF (Volume Fraction) method [1, 2], homogenization is achieved at the level of the difference cell of a grid and the degree of homogenization can always be decreased to an acceptable value depending on the required computational accuracy and resources.

Matrix multiplication is one of the core operations in many areas of scientific computing. We present the results of the experiments with the matrix multiplication of the big size comparable with the big size of the onboard memory, which is 1.5 terabyte in our case. We run experiments on the computing board with two sockets and with two Intel Xeon Platinum 8164 processors, each with 26 cores and with multi-threading. The most interesting result of our study is the observation of the perfect scalability law of the matrix multiplication, and of the universality of this law.

We report on the testing of write operation in cloud for OpenNebula package. Results obtained for writing small blocks with zero delay show stochastic behaviour, which could be explained by compound influence of optimization technologies, such as caching, used in hypervisor.