Illuminating the I/O Optimization Path of Scientific Applications

The existing parallel I/O stack is complex and difficult to tune due to the interplay of multiple factors that impact the performance of data movement between storage and compute systems. When performance is slower than expected, end-users, developers, and system administrators rely on I/O profiling and tracing information to pinpoint the root causes of inefficiencies. Despite having numerous tools that collect I/O metrics on production systems, it is not obvious where the I/O bottlenecks are (unless one is an I/O expert), their root causes, and what to do to solve them. Hence, there is a gap between the currently available metrics, the issues they represent, and the application of solutions and optimizations that would mitigate slowdowns. An I/O specialist often checks for common problems before diving into the specifics of each application and workload. Streamlining such analysis, investigation, and recommendations could close this gap without requiring a specialist to intervene in each case. In this paper, we propose a novel interactive, user-oriented visualization and analysis framework to pinpoint various root causes of I/O performance problems and provide a set of actionable recommendations to improve performance based on the observed characteristics of an application. We evaluate its applicability and correctness in four use cases from distinct science domains and demonstrate its value to end-users, developers, and system administrators when seeking to improve an application's I/O performance.

---

Contributors:

• Hammad Ather
• Boyana Norris
• Jean Luca Bez