(PP09) ProfiT-HPC: A Profiling Toolkit for HPC in Tiers 2 and 3
Performance Analysis and Optimization
TimeTuesday, June 26th3:15pm - 3:45pm
DescriptionProfiT-HPC aims at raising the awareness of HPC users in all fields of research for the importance of profiling and performance measurements of their applications. The primary goal of Profit-HPC is to lower the barrier to the efficient usage of HPC systems. For this, a toolkit will be developed that automatically collects monitoring data and delivers an easy to digest summarized text report to the HPC users containing a summary of resources used by the job. Combining the batch system information of requested resources with the collected metrics, the user gets suggestions on how to schedule the job with a more efficient resource usage. Through interactive dashboards, the user can zoom into different job-specific metric detail levels. Additionally, a system wide administrator view is currently developed providing administrators of data centers an easier access to the system state as well as interaction with the users on a job-specific basis. All collected information should help to turn the attention of even those users with little HPC background knowledge to the possible inefficiencies or even problems and in a longer term, to improve the general quality of the used applications. The next steps of the project will be to include performance profiling data in the automatically generated reports to identify often encountered issues and typical patterns of sub-optimal program behavior.