Score-P measurement system code and event logs for Euro-Par 2018 paper: A Methodology for Performance Analysis of Applications Using Multi-layer I/O

This artifact contains the prototype implementation of a novel approach for recording I/O operations across multi-layer I/O interfaces and a case study using that approach, as described in the accompanying paper, "A Methodology for Performance Analysis of Applications Using Multi-layer I/O".<br>In order to fully utilise the high bandwidth offered by modern high-performance computing storage systems, and to handle increasingly large volumes of data, parallel I/O operations are often implemented through libraries such as HDF5, NetCDF and MPI I/O. In many cases, multiple libraries may operate concurrently within a given software stack. This concurrency and operations between libraries can impact upon each other. Given this, it is essential to be able to effectively monitor I/O operations across all libraries utilised by a given application.<br>Here, a prototype methodology for recording calls to I/O libraries on multiple layers of the software stack is presented. This methodology is implemented as an extension of Open Trace Format Version 2 (OTF2) and the Score-P measurement infrasctructure. The implementation is provided here in the tarball <b>scorep-4.0-io.tar.gz</b><b><br></b>In addition to the implementation, event logs generated by its test evaluation of the Met Office NERC Cloud model (MONC) simulator are provided. Therein, the I/O behavior of MONC using the Score-P implementation are provided. These logs, provided in the directory <b>monc_112n_14N_10T_small</b>, can be visualized in a tool such as Vampir.<br>Further instructions on recreating the analyses presented in the accompanying paper are provided in <b>artifact.pdf</b>.