Аннотация:In order for supercomputer resources to be effectively used, it is necessary to constantly analyze various aspects of the operation of modern HPC systems. One of the most significant aspects is the efficiency of execution of parallel applications running on a supercomputer. To study this, system administrators need to constantly monitor and analyze the entire flow of running jobs. This is a very difficult task, and there are several reasons for this - a large number and a significant variety of executed applications; the extreme complexity of the structure of modern HPC systems, and, as a result, a huge number of characteristics that need to be evaluated for each job. One way to make this analysis easier is to cluster similar jobs. Such clustering allows you to infer the behavior and performance issues of all jobs in the cluster by examining only one of these jobs, and it also helps to better understand the structure of the supercomputer job flow as a whole. In this paper, we propose a new method that allows solving this clustering task with high accuracy. This smart clustering method analyzes both static information on the executable files and dynamic data about the behavior of applications during their execution. Using the Lomonosov-2 supercomputer as an example, we demonstrate how this method can help in practice to facilitate the analysis of the execution efficiency of supercomputing applications.