Tim Robinson, a staff member of CSCS, was awarded best paper at the recent Cray User Group (CUG) meeting in Stuttgart. The paper, entitled “Software usage on Cray systems across three centers (NICS, ORNL and CSCS)”, was co-authored with collaborators at Oak Ridge National Laboratory and the National Institute for Computational Sciences, Tennessee.
The paper describes the implementation of a software tool to automatically and transparently determine the usage of numerical libraries, compilers and applications, right down to the level of specific version numbers. The approach is based on intercepting the GNU linker – to gather information on libraries linked against binaries – and the MPI job launcher – to monitor the execution of those binaries. Although developed specifically for Cray systems, the tool is readily transferable to other high performance computing clusters.
The tool allows HPC centres to monitor software usage and forecast needs – for example, to determine which compiler suites are most used and should continue to be supported in the future. Moreover, the tool can alert support staff to situations where a user is running code that is linked to legacy or deprecated libraries, or even worse, libraries known to be buggy.