Using IOR to Analyze the I/O Performance for HPC Platforms
Hongzhang Shan, John M. Shalf · University of North Texas Digital Library (University of North Texas) · 2007
The HPC community is preparing to deploy petaflop-scale computing platforms that may include hundreds of thousands to millions of computational cores over the next 3 years.Such explosive growth in concurrency creates daunting challenges for the design and implementation of the I/O system.In this work, we first analyzed the I/O practices and requirements of current HPC applications and used them as criteria to select a subset of microbenchmarks that reflect the workload requirements.Our analysis led to selection of IOR, an I/O benchmark developed by LLNL for the ASCI Purple procurement, as our tool to study the I/O performance on two HPC platforms.We selected parameterizations for IOR that match the requirements of key I/O intensive applications to assess its fidelity in reproducing their performance characteristics.