A taxonomy of how the input/output behavior of applications on high-performance computing systems is measured, analyzed, and improved. Learners study the layered HPC I/O stack (high-level libraries, MPI-IO and POSIX interfaces, I/O forwarding, parallel file systems, and storage hardware), the profiling and tracing tools that reveal I/O bottlenecks, statistical, visual, and machine-learning methods for characterizing I/O, and the optimization strategies (collective I/O, request aggregation, scheduling, striping) that raise achieved bandwidth.
A survey (submitted to ACM Computing Surveys) of the tools and techniques used to characterize, analyze, and optimize the input/output behavior of applications on large-scale high-performance computi…