Skip to main content

IOR

Parallel I/O benchmark suite for evaluating HPC storage system performance at scale.

Overview​

IOR is the de-facto standard for benchmarking parallel I/O performance in HPC systems. It provides comprehensive testing of read/write performance across different I/O patterns and storage backends, making it essential for HPC storage evaluation and tuning.

Key Features​

  • Comprehensive Benchmarking: Multiple I/O patterns (sequential, random, collective)
  • Scale Testing: Test performance from single node to thousands of nodes
  • Multiple Backends: Support for POSIX, HDF5, NetCDF, and other I/O libraries
  • Performance Analysis: Detailed metrics for bandwidth, latency, and efficiency
  • Reproducible Results: Standardized testing methodology for fair comparisons

Technology Stack​

  • Language: C
  • Tags: HPC, Benchmarking, I/O Performance, Storage Systems, Analysis
  • Source Organization: HPC Community

Repository​

📦 GitHub: hpc/ior

Use Cases​

  • Evaluating HPC storage system performance
  • Comparing different storage backends and configurations
  • Validating I/O optimization strategies
  • Characterizing application I/O behavior
  • Capacity planning for HPC infrastructure

Running IOR​

IOR supports various command-line options for different workload patterns:

# Simple sequential write benchmark
mpirun -np 64 ior -w -n 1G

# Random read pattern
mpirun -np 64 ior -r -s 10m -b 1m

# Collective I/O operations
mpirun -np 64 ior -c -t 4m -b 256m

Getting Started​

Visit the IOR GitHub repository for installation instructions, comprehensive documentation, and advanced usage examples.

Publications & Community​

IOR has been used in hundreds of HPC storage studies and is widely cited in performance analysis literature. Active community contributions welcome.

Status​

Production - Widely adopted, actively maintained by HPC community