RUS  ENG
Full version
JOURNALS // Vestnik Yuzhno-Ural'skogo Gosudarstvennogo Universiteta. Seriya "Vychislitelnaya Matematika i Informatika" // Archive

Vestn. YuUrGU. Ser. Vych. Matem. Inform., 2024 Volume 13, Issue 3, Pages 61–78 (Mi vyurv322)

TASC software for HPC performance analysis: current state and latest developments

Vad. V. Voevodin, D. I. Shaikhislamov, V. A. Serov

Lomonosov Moscow State University (Leninskie Gory 1, Moscow, 119991 Russia)

Abstract: To ensure high operating efficiency of modern supercomputers, it is necessary to constantly analyze and control various aspects of their behavior, paying special attention to the flow of supercomputer applications running on these machines. To solve this problem, the TASC (Tuning Applications for SuperComputers) software suite was previously developed. It automatically detects performance issues in HPC applications and evaluates the efficiency of using supercomputer resources, provides supercomputer administrators with a flexible report tool for analyzing different aspects of supercomputer functioning with the desired level of detail, and estimates the noise level on compute nodes. This paper provides full-scale description of current TASC structure and capabilities, including the stages of data processing and storing, as well as performing different types of analysis. It also describes new results obtained and methods developed within one of the main TASC components — assessment system for quick and accurate evaluation of HPC resources usage efficiency.

Keywords: high-performance computing, supercomputer, performance analysis, workload analysis, operational data analytics, monitoring.

UDC: 519.68

Received: 04.09.2024

DOI: 10.14529/cmse240304



© Steklov Math. Inst. of RAS, 2024