Development of Methods for Dispatching User Renewable Threads in Multi-core Systems

The work is aimed to improve the efficiency of solving user tasks in dynamic high-performance computing systems by modifying dispatching algorithms. A formalized representation of the task of dispatching user renewable threads in high-performance systems, as well as algorithms for dispatching user renewable threads in high-performance systems was proposed. As a result of the statistical data analysis of the first test, the following features of the developed algorithm were revealed in comparison with the best results of existing algorithms: average delay reduction by 4.94%; an increase in delays by an average of 3.69% (from 0% to 5%) in 13% of tasks; delays are reduced by an average of 18% (from 15% to 23%) in 8% of tasks; delays are reduced by an average of 6.3% (from 0% to 14%) in 79% of tasks. The results of the work are applicable in applications requiring high efficiency of computing resources.