The SPP benchmarks were originally used as part of the Blue Waters acceptance test and were updated in 2017 to represent the change in workloads. The benchmarks and methodolgy have not been updated since 2017.
The following pages contain links and documentation for to several benchmark related collections used for system performance assessment within the Blue Waters project. There are three main types of bencmark tests that are iintended to effectively evaluate the performance of a system with a leadership class workload reflecting the open research workload that runs on the Blue Waters system. A detailed report (download, support) of the workload running on Blue Waters for the first three years of production full services and done in collaboration with the XDMoD CCR group at SUNY, is availalbe.
The Sustained Petscale Performance benchmark is represenative of the workload observed on Blue Waters over the first 3 years of operation. The I/O benchmarks are a collection of standards plus two applications. The low level benchmarks are industry standards such as HPCC, NPB, etc. Instructions for obtaining the benchmarks and rules for reporting are provided for each section.
The SPP benchmarks include a description of the SPP Methodology and Evaluation tables and then a separate page for the actual codes, input problems and run instructions are on the SPP Code and Input page.
Application based I/O benchmarks and synthetic I/O benchmarks are on the I/O Benchmarks page.
Synthetic benchmarks such as NPB, STREAM etc are on the Low Level Benchmark Instructions and Description page.
Please submit issues regarding the benchmarks to help+bw@ncsa.illinois.edu.
