Analysis and optimization of quantitative characteristics of the formation of multi-threaded parallelism processes in BigData sorting tasks

A set of software and hardware totality for solving the problem of increasing the speed of processing big data through parallelization by organizing multi-threading is considered and analyzed. The solution of this problem is carried out in the aspect of the problem of increasing the productivity of processing big data for artificial intelligence and machine learning algorithms. Numerical modeling and analysis of processor time consumption when organizing multithreading were performed on examples of array sorting for matrices of different dimensions and trends in changes in the optimal number of threads were analyzed. The hypothesis of a monotonic (quasi-monotonic) dependence of the reduction of processor operating time with an increase in the number of threads was numerically verified. A comparative analysis of the implementation of parallel computing was carried out on hardware of different configurations and using different programming languages.