Mango-IO: I/O Metrics Consistency Analysis

IEEE International Conference on Cluster Computing Workshops (CLUSTER Workshops), pp. 18–24

Abstract

Performance tools are inseparable from complex HPC applications’ performance analysis and engineering life cycles. Due to the application’s complexity, various performance analysis tools are created to serve different analysis purposes and provide a deeper look at certain aspects of the applications. Although these tools might operate differently, having coherent information and consistent metrics across all tools is mandatory for ensuring analysis continuity. It is common for performance analysts to switch their usual performance tools due to various reasons and limitations. In this work, we look specifically at the I/O performance analysis tools landscape and introduce Mango-IO to verify the result consistencies between tools and provide tool-agnostic metrics calculation methods. Our analysis and case study provides lesson learned and guideline for ensuring measurement continuity and comparability.

Authors 4

1. Radita Liem Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Aachen,Germany

   RWTH Aachen University, Aachen, Germany

2. Sebastian Oeste

   Technische Universität Dresden

   Affiliation as printed

   TU Dresden,Dresden,Germany

   TU Dresden, Dresden, Germany

3. Jay Lofstead

   Sandia National Laboratories

   Affiliation as printed

   Sandia National Labs,Albuquerque,NM,USA

   Sandia National Labs, Albuquerque, NM, USA

4. Julian Martin Kunkel

   Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen · University of Göttingen

   Affiliation as printed

   Göttingen University/GWDG,Göttingen,Germany