

StressBench: A Configurable Full System Network and I/O Benchmark Framework
Only 126 seats left
Tuesday, June 29, 2021 3:00 PM to 4:00 PM · 1 hr. (Africa/Abidjan)
Information
Contributors:
Abstract:
This poster demonstrates StressBench, a network and I/O benchmark framework designed to test the network and I/O subsystems with representative application communication patterns and I/O traffic. We first demonstrate how StressBench can used as a microbenchmark suite similar to the Intel MPI Benchmarks, and the OSU Microbenchmarks. We then validate application replay through StressBench with Sweep3D demonstrating less than 6% difference between the real application runs and the emulated job. Finally, we build on the application replay functionality to demonstrate how StressBench can be used to replicate a full system workload, demonstrating the effect of performance degradation due to applications contending for the network and I/O resources.
- Gihan Mudalige (University of Warwick)
- Steven Wright (University of York)
- Dean Chester (University of Warwick)
Abstract:
This poster demonstrates StressBench, a network and I/O benchmark framework designed to test the network and I/O subsystems with representative application communication patterns and I/O traffic. We first demonstrate how StressBench can used as a microbenchmark suite similar to the Intel MPI Benchmarks, and the OSU Microbenchmarks. We then validate application replay through StressBench with Sweep3D demonstrating less than 6% difference between the real application runs and the emulated job. Finally, we build on the application replay functionality to demonstrate how StressBench can be used to replicate a full system workload, demonstrating the effect of performance degradation due to applications contending for the network and I/O resources.