PIONEER: A Solution to Parallel I/O Workload Characterization and Generation

The demand for parallel I/O performance continues to grow. However, modelling and generating parallel I/O work-loads are challenging for several reasons including the large number of processes, I/O request dependencies and workload scalability. In this paper, we propose the PIONEER, a complete solution to Parallel I/O workload characterization and gEnERation. The core of PIONEER is a proposed generic workload path, which is essentially an abstract and dense representation of the parallel I/O patterns for all processes in a High Performance Computing (HPC) application. The generic workload path can be built via exploring the inter-processes correlations, I/O dependencies as well as file open session properties. We demonstrate the effectiveness of PIONEER by faithfully generating synthetic workloads for two popular HPC benchmarks and one real HPC application.