odin-data-dpdk

DPDK-based acceleration libraries for odin-data

Overview

This repository extends odin-data with DPDK-accelerated frame processing capabilities. It provides:

• DPDK Frame Processor Plugin: High-performance frame processing using DPDK worker cores
• Worker Cores: Modular processing pipeline including:
  • Packet reception cores
  • Packet processing cores
  • Frame building cores
  • Frame compression cores
  • Frame wrapper cores
  • Python access cores
• Network and Camera Support: Protocols and decoders for various detector types
• TensorStore Integration: Optional support for Tensorstore high speed writing

Repository Structure

odin-data-dpdk/
├── cpp/                     # C++ source code
│   ├── src/                 # Core implementations
│   ├── include/             # Public headers
│   ├── config/              # Configuration files
│   └── CMakeLists.txt       # Build configuration
├── docs/                    # Documentation
│   ├── installation.md      # Build and installation guide
│   └── configuration.md     # Configuration examples
├── performance_testing/     # Performance testing tools
└── tools/                   # Utility scripts

Prerequisites

• CMake >= 3.24
• Python >= 3.10
• GCC >= 10
• G++ >= 10
• DPDK
• odin-data (must be installed first)

Quick Start

Basic Installation

# Clone and build odin-data first
cd ~/develop/repos
git clone https://github.com/odin-detector/odin-data
cd odin-data/cpp
git checkout 1.10.2
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=~/develop/install ..
make -j install

# Clone and build odin-data-dpdk
cd ~/develop/repos
git clone https://github.com/stfc-aeg/odin-data-dpdk
cd odin-data-dpdk/cpp
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=/develop/install -DODINDATA_ROOT_DIR=/develop/install -DCMAKE_MODULE_PATH=/develop/odin-data/cpp/cmake ..
make -j install

Environment Setup

On our current systems running Ubuntu 22.04.5, only CMake needs to be loaded (Python 3.10+ and GCC 10+ are available by default):

module load cmake/3-31-6

On older Ubuntu versions or systems configured differently, you may also need to load Python and GCC e.g.

module load python/3-11-7
module load cmake/3-31-6
export CC=/usr/bin/gcc-10
export CXX=/usr/bin/g++-10
# or whatever versions you have avalible that meet the minimum requirements

To check what version of these you have available use module avail

Usage

Run the frame processor with DPDK plugin:

cd ~/develop/install
module load dpdk/23.11.5
./bin/frameProcessor --ctrl tcp://0.0.0.0:5000 --log-config config/fp_log4cxx.xml --config config/<config_file_name>.json

Available Plugins

Detector specific plugins can built on top of this folder structure to allow odin-data-dpdk to load them at runtime. See available plugins below:

Packet-Based Plugins

• mercury-detector
• babyd-detector

Camera-Based Plugins

• odin-orca-quest

Tensorstore Support

Tensorstore support can be enabled at build time:

cmake -DCMAKE_INSTALL_PREFIX=/develop/install -DODINDATA_ROOT_DIR=/develop/install -DENABLE_TENSORSTORE=ON -DCMAKE_MODULE_PATH=/develop/odin-data/cpp/cmake ..

NOTE: The fake cameras used to pass frames into the Tensorstore Core were modifed to enable them to output frames faster. These changes have not beenm commited as to not break other work that may be done with this repo.

Python Tools

The tools/python/ directory contains several utilities for interacting with and analyzing odin-data-dpdk:

GUI and Control Tools

odin-gui.py

A GUI for controlling and monitoring odin-data-dpdk frame processors via ZMQ

python tools/python/odin-gui.py

liveviewer.py

Real-time frame visualization tool that connects to the frame processor's ZMQ stream

python tools/python/liveviewer.py --endpoint tcp://localhost:5555

Data Analysis Tools

plot_tensorstore_performance.py

Analyzes and visualizes Tensorstore write performance from CSV logs using Matplotlib

python tools/python/plot_tensorstore_performance.py
# Opens file dialog to select CSV performance logs

TensorstoreViewer.py

Interactive viewer for Tensorstore datasets (Zarr2/Zarr3)

# View a Zarr dataset
python tools/python/TensorstoreViewer.py /path/to/dataset

# View Zarr3 dataset
python tools/python/TensorstoreViewer.py /path/to/dataset --spec zarr3

Development Tools

frame_producer

UDP frame producer for testing and simulation. Useful for testing frame reception without a physical detector.

bifrost

Python library for zero-copy frame processing from odin-data-dpdk

See bifrost/README.md for detailed usage.

Installation

Install Python tool dependencies (preferably in a virtual environment):

cd tools/python
pip install -r requirements.txt

Documentation

• Configuration Guide