Разработка комплекса параллельных программ решения обратных задач гравиметрии и магнитометрии для сеток большой размерности

Abstract

The most important problems in studying of the structure of the earth’s crust are the inverse problems of gravimetry and the problems of magnetometry. The problem is in finding interfaces between layers with different constant densities using known gravitational data. The methods for solving these problems are based on the ideas of iterative regularization. After discretization, these are reduced to systems of nonlinear functions of large dimensions. The need to improve the accuracy of the results of solving problems entails an increase in the computation time. The paper describes a remote computing system and an integrated program complex for graphics accelerators that implement the fastest and most memory-efficient algorithms developed earlier and based on gradient methods. The remote computing system considered in the work is a web portal that is a universal solution for launching tasks on remote clusters. The most important advantage of the portal is the simplicity of its use: when connecting to a cluster to carry out calculations, you no longer need to install additional software on the cluster itself, and you do not need a privileged account to work with the cluster. All that is required is a valid account on the remote cluster, the rest of the work on communication with the data processing center (DPC) is taken over by the portal. The portal can easily scale with the growth of the number of users who can download the necessary algorithms and perform calculations using the data center.