Laboratory for analysis of supercomputer systems and applications

The main focus of the laboratory is a comprehensive analysis of the quality of modern supercomputer functioning in order to increase performance as well as ensure operability and support of such systems. The extreme complexity of supercomputer architecture leads to the fact that even administrators of such systems do not have a sufficient understanding of what processes occur during their operation and how this affects the execution of the flow of user applications. The situation is complicated by the fact that modern supercomputers most often run a huge variety of applications from different subject areas, which have diverse properties and for the effective implementation of which different factors must be considered. All this leads to a decrease in both overall performance and the operability of supercomputers, as a result of which a significant part of the computational resources is idle. The activities of this laboratory are aimed at solving this global problem.

The following main tasks of the laboratory can be distinguished:

• development of new methods for analyzing the performance of supercomputer applications, both through a detailed study of individual programs and through continuous study of the entire flow of running applications;

• “intellectual” analysis of large volumes of data on the quality of operation of a supercomputer and the applications running on it, which allows obtaining new knowledge and create new methods for studying the properties of applications and the behavior of supercomputer components;

• development and implementation of approaches to improve the reliability and efficiency of supercomputer functioning;

• application of supercomputer co-design technologies to match the properties of applications with the characteristics of hardware and system software;

• analysis and optimization of the performance of real user applications running on existing supercomputer systems.