Analysis of data on the loading of high-performance platforms by user tasks on the example of the heterogeneous computing platform HybriLIT

Abstract

The relevance of the research presented in this article lies in the ability to aggregate statistical information about the use of platform resources by various user groups. The article analyzes the data on the use of the resources of the training and testing ground of the HybriLIT platform, on the basis of which it is possible to build models for predicting the further workload of the platform in order to rationally allocate the available computing resources and the data storage system, as well as increase the efficiency of their use. The purpose of the work is to present the research aimed at data mining of the resources used when launching tasks by various groups of users and the time of their execution on the training and testing ground of the HybrLIT platform. The heterogeneous HybriLIT computing platform, consisting of a training and testing ground and the Govorun supercomputer, is part of the multifunctional information and computing complex of the Meshcheryakov Information Technology Laboratory of the Joint Institute for Nuclear Research. The platform has a heterogeneous structure of computing nodes and allows you to run parallel applications for performing calculations on various computing architectures. The summary information about the tasks running on HybriLIT, which is automatically recorded in the database by the SLURM resource manager and scheduler, is of interest for analysis. As a result, a deep analysis of data was carried out according to several criteria for each resource and for each group of users of the training and testing ground of the heterogeneous HybriLIT platform.