Evaluation of effectiveness of high performance computing system with functional models

Educational institutions, organizations and companies operating in knowledge-intensive industries, are often in need to make conducting intensive calculations using the high-performance supercomputing systems and computer networks. As the algorithms of distribution of tasks within such systems, people use standard algorithms aimed at efficient operation with a lot of similar tasks. At the meantime, there is a whole class of problems, which requires the successful completion of a number of conditions, such as having certain problem-oriented software, specific requirements for hardware resources of compute nodes, and others. This class of tasks also includes the tasks associated with the calculation of the path of movement of the planets, geological exploration, big-data analysis, etc. In this case, for the successful distribution of the load within a high-performance system we use special algorithms for planning and allocation of tasks. So, effective usage of computing resources in case of high-performance computing centers is the actual problem. The main aspect of this problem is the approach of task planning and distribution across computing nodes. This article covers the main algorithms of huge tasks distribution in large computing systems, such as First Come First Served, Shortest/Longest Job First, Backfilling, Round-robin, etc. To evaluate the effectiveness of these algorithms the authors have developed a model of a computer system, showing the structure of an existing computer system of SFU. As the initial experimental data authors used a variety of tasks which have run in the computing system of SFU over the past few years. As a platform to simulate running tasks SimGrid platform and Alea were used. The closest to the real conditions were achieved for the evaluation of the effectiveness of different algorithms of distribution of user tasks. Using the experiments results, the authors propose some solutions for the modernization of the existing computing infrastructure.