User GuideBenchmarking NIXLNIXLBench

Building NIXLBench

View as Markdown

NIXLBench requires a NIXL installation — see Building NIXL from Source for instructions.

System Requirements

Hardware

  • CPU: x86_64 or aarch64 architecture
  • Memory: 8 GB RAM minimum, 16 GB or more recommended for compilation
  • Storage: 20 GB free disk space
  • GPU: NVIDIA GPU with CUDA support (for GPU features)
  • Network: InfiniBand or Ethernet adapters (for network backends)

Software

  • OS: Ubuntu 22.04 or 24.04 LTS (recommended), or RHEL-based distributions
  • Docker: 20.10 or later (for container builds)
  • CUDA Toolkit: 12.8 or later
  • Python: 3.12 or later
  • Compiler: C++20-compatible compiler

Build Instructions

The Docker build handles all dependencies automatically and is the recommended approach.

Clone the repository and navigate to the build directory:

git clone https://github.com/ai-dynamo/nixl.git
cd nixl/benchmark/nixlbench/contrib

Build the container with default settings:

./build.sh

Common build options:

# Debug build
./build.sh --build-type debug
# Build for aarch64
./build.sh --arch aarch64
# Combine options
./build.sh --build-type debug --arch aarch64

For the full list of build options, see the NIXLBench README.

Run a quick benchmark to verify the build:

docker run -it --gpus all --network host nixlbench:latest \
  nixlbench --backend GDS --filepath /tmp/nixlbench-test