Challenges in Understanding Metadata Performance: A Case of Metadata Analysis Using Score-P

IEEE International Conference on Cluster Computing Workshops (CLUSTER Workshops), pp. 113–117

Abstract

Modern scientific applications are often data-intensive due to the use of AI. A substantial portion of the I/O operations in these applications comes from metadata operations. Analyzing the performance of metadata operations is the key to improving the application's performance. In this work-in-progress paper, we are looking into the role played by metadata, the current status quo of the tool, and the benchmark to assess metadata performance from the initial results we collected using Score-P on MDWorkbench. Preliminary findings show the complexity of understanding the metadata performance and the challenges in handling outlier performance data.

Authors 2

1. Boris Kosmynin Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Aachen,Germany

2. Radita Liem Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University,Aachen,Germany