RUS  ENG
Full version
JOURNALS // Numerical methods and programming // Archive

Num. Meth. Prog., 2023 Volume 24, Issue 1, Pages 24–36 (Mi vmp1072)

Parallel software tools and technologies

Development of a portable software solution for monitoring and analyzing the performance of supercomputer applications

Vadim V. Voevodin, Konstantin S. Stefanov

Lomonosov Moscow State University, Research Computing Center, Moscow, Russia

Abstract: Modern supercomputers are used in various areas of science and technology. However, their computational resources are often not fully utilized. The reason often lies in the low efficiency of user applications. However, it is very difficult to solve this problem, which is due both to the extreme complexity of the structure of modern supercomputers, and to the lack of theoretical knowledge and practical experience in creating highly efficient parallel applications among users of computing systems. Moreover, users are often not even aware that their applications are not working efficiently. Therefore, it is important for supercomputer administrators to be able to constantly monitor and analyze the entire flow of running applications. For these purposes, different existing systems for monitoring and analyzing performance can be used, however, most of such solutions do not provide sufficient functionality for studying performance, or are not portable. This paper describes a prototype of the software package being developed, which provides wide opportunities for collecting and automatically analyzing application performance data and is portable at the same time.

Keywords: parallel computing, supercomputer, monitoring, data analysis, performance, portability.

UDC: 519.68

Received: 07.11.2022
Accepted: 22.02.2022

DOI: 10.26089/NumMet.v24r103



© Steklov Math. Inst. of RAS, 2024