Skip to content

uni-tue-kn/P4TG

BranchesTags

Repository files navigation

P4TG Logo

P4TG: Traffic Generation for Ethernet/IP Networks

License Language Built with P4 Version Controller Build Data Plane Build

📑 Table of Contents

📝 Overview

P4TG (P4-based Traffic Generator) is a programmable traffic generator built on the Intel Tofino™ ASIC.
It enables reproducible experiments with Ethernet/IP traffic at high data rates while providing precise in-data-plane measurements.
P4TG combines a P4 data plane program, a Rust-based control plane, and a React-based configuration GUI.

P4TG Architecture

📖 Related publications:

Traffic Generation

  • Tofino 1: Up to 1 Tb/s across 10 × 100 Gb/s ports, or 40 × 10/25 Gb/s
  • Tofino 2: Up to 4 Tb/s across 10 × 400 Gb/s ports, or 40 × 10/25/100 Gb/s, or 80 × 10/25/50 Gb/s in channelized mode

Generated packet headers can be fully customized, including:

  • Ethernet
  • IPv4 (full address randomization)
  • IPv6 (address randomization of 48 LSB)
  • MPLS (up to 15 LSEs)
  • VLAN
  • QinQ
  • VxLAN
  • SRv6 (up to 3 SIDs, Tofino2 only)

Statistics

Traffic may be looped back (directly or through other equipment) to measure:

  • TX/RX rates
  • Packet loss
  • Packet reordering
  • Inter-arrival times (IATs)
  • Round-trip times (RTTs)
  • Frame types
  • Frame sizes

directly in the data plane to ensure accurate results.

Version Highlights

P4TG Architecture

See the full Changelog.

Features

Feature Status
Statistics ✅ Available
Tofino 1 & 2 support ✅ Available
Web frontend (React) ✅ Available
Rust backend ✅ Available
Ethernet, IPv4, IPv6 ✅ Available
VLAN, QinQ, MPLS, VxLAN, GTP-U, SRv6 ✅ Available
ARP replies ✅ Available
Dark mode ✅ Available
IAT+RTT histogram & percentiles ✅ Available
Automated testing ✅ Available
Line rate traffic patterns (e.g., flashcrowd) ✅ Available
File reporting ✅ Available
Test profiles ⚠️ Experimental
Localization ⚠️ Experimental
NDP ⏳ Planned
NETCONF ⏳ Planned

🚀 Installation & Quick Start

P4TG requires a fully set up bf-SDE with $SDE and $SDE_INSTALL environment variables set. A detailed installation guide for the Intel SDE and P4TG can be found here.

Quick Start

The provided p4tg.sh script automates the installation of the data and control plane on Debian- / Ubuntu-based systems, provided that the SDE is installed correctly. Make sure that the environment variables $SDE and $SDE_INSTALL are set. Run with sudo -E to pass environment variables.

Usage: sudo -E ./p4tg.sh [install|update|start|stop|restart|status][--nightly]

Clone P4TG into /opt/P4TG and simply run sudo -E ./p4tg.sh install (tested on Debian-based systems). Change the paths at the top of p4tg.sh if needed.

The install command will:

  • Compile the data plane and copy it in place.
  • Pull the docker image and start it.
  • Copy the p4tg.sh script to /usr/local/bin.
  • Copy the p4tg.service file to /etc/systemd/system/p4tg.service. This service file can be used to autostart P4TG on boot.

The start command will:

  • Load all required kernel modules to operate the Tofino.
  • Start the data plane and wait for it to become ready.
  • Start the control plane docker image.

The control plane docker image:

  • Starts a REST-API server on port P4TG_PORT (default: 8000) at /api
  • Serves the React GUI at /
  • Access GUI: http://<tofino-controller-ip>:P4TG_PORT
  • Access API: http://<tofino-controller-ip>:P4TG_PORT/api

Configuration

Docker compose file Controller/docker-compose.yaml

  • SAMPLE=1 → IAT sampling mode (default: 0, data plane measurement)
  • LOOPBACK=true → enable loopback testing mode
  • P4TG_PORT=8000 → changes the controller port
  • NUM_PORTS=32 → set number of front panel ports of your device

Config file: Controller/config.json

  • Specify ports for traffic generation. Per default, front panel ports 1 - 10 are configured automatically.
  • Configure MAC address for ARP replies, channel_count, and port settings. Port settings can further be changed during runtime through the API or the GUI.

Example:

{
  "tg_ports": [
    {
      "port": 1,
      "mac": "00:d0:67:a2:a9:42",
      "speed": "BF_SPEED_100G"
    },    
    {
      "port": 2,
      "mac": "fa:a6:68:e0:3d:70",
      "channel_count": 4,
      "speed": "BF_SPEED_25G"
    },
    {
      "port": 4,
      "mac": "d6:67:75:a1:94:c3",
      "channel_count": 8,
      "speed": "BF_SPEED_50G"
    },
    {
      "port": 3,
      "mac": "00:d0:67:a2:a9:42",
      "speed": "BF_SPEED_100G",
      "fec": "BF_FEC_TYP_NONE",
      "auto_negotiation": "PM_AN_FORCE_DISABLE"
    }
  ]
}

Configuration Options

Option Valid Values
mac Any valid MAC address
speed BF_SPEED_10G · BF_SPEED_25G · BF_SPEED_40G · BF_SPEED_50G · BF_SPEED_100G · BF_SPEED_400G
channel_count 4 · 8
auto_negotiation PM_AN_DEFAULT · PM_AN_FORCE_ENABLE · PM_AN_FORCE_DISABLE
fec BF_FEC_TYP_NONE · BF_FEC_TYP_FC · BF_FEC_TYP_REED_SOLOMON
breakout_mode Deprecated: true · false

Notes:

  • speed always describes the per-channel speed. Without channel_count, the port is configured as 1x<speed>.
  • Valid channel_count combinations are:
    • Tofino 1: 4x10G, 4x25G
    • Tofino 2: 4x10G, 4x25G, 4x100G, 8x10G, 8x25G, 8x50G
  • channel_count: 4 with speed: 100G on Tofino 2 uses channels 0,2,4,6.
  • channel_count: 4 with speed: 10G/25G uses channels 0,1,2,3.
  • Runtime speed changes through the GUI or POST /api/ports are rejected if they would require a different active channel layout. For example, 4x25G -> 4x100G requires updating config.json and restarting the controller.
  • Backward compatibility: breakout_mode: true is deprecated, logs a warning, and is interpreted as legacy 4-channel breakout. breakout_mode: false is deprecated, logs a warning, and disables channelization.
  • ARP reply and MAC can be changed at runtime per port/channel in the Ports GUI or via POST /api/ports/arp (optional channel field).
  • Runtime ARP/MAC changes are kept in controller memory and are reset to config.json values on controller restart.
  • Default/mandatory FEC rules:
    • 400G, 4x100G, and 8x50G use BF_FEC_TYP_REED_SOLOMON
    • 4x10G, 4x25G, 1x10G, 1x25G, 1x40G, 1x50G, and 1x100G default to BF_FEC_TYP_NONE
    • 1x50G additionally allows BF_FEC_TYP_REED_SOLOMON to be configured manually if needed

64-port Tofino

Each front panel port requires two recirculation ports. They are configured automatically. However, on the 64-port Tofino devices, you may have to manually set internal TX/RX recirculation ports to match your device. Below is an example:

{
  "tg_ports": [
    {
      "port": 49,
      "mac": "fa:a6:68:e0:3d:70",
      "speed": "BF_SPEED_100G",
      "recirculation_ports": {
        "tx_port": 50,
        "rx_port": 51
      }
    }
  ]
}

This uses front panel port 49 as a port for traffic generation and port 50 and 51 for internal recirculation. Ensure that the recirculation ports support the same line rate as the TG port.

🤖 Test Automation

Automation scripts in p4tg_test_automation:

  • Python script for starting/stopping tests, fetching stats, and rendering plots
  • Quick start: 
    python run.py --payload payloads/your_test.json
  • Example payloads included

See README for details.

🔄 Update Guide

Run sudo -E p4tg.sh update.

Manually

  1. Rebuild the data plane as described here.
  2. Update controller: 
    docker compose pull
docker compose up -d
  3. If GUI issues occur, clear browser storage.

📚 Documentation

🛠️ Development

For development instructions, please see here

🖼️ Preview

Preview 1

Preview 2

Preview 3

Preview 4

Preview 5

🏢 Who's Using P4TG

Airbus     BDBOS     Bell     BelWü     ECI

... and many more!

📖 Cite

If you use P4TG in any of your publications, please cite the following papers:

  1. S. Lindner, Marco Häberle, and M. Menth: P4TG: 1 Tb/s Traffic Generation for Ethernet/IP Networks, in IEEE Access, vol. 11, p. 17525 – 17535, Feb. 2023, IEEE
  2. F. Ihle, E. Zink, S. Lindner, and M. Menth: Enhancements to P4TG: Protocols, Performance, and Automation, in KuVS Workshop on Network Softwarization (KuVS NetSoft), online, Apr. 2025
  3. F. Ihle, E. Zink, M. Menth: Enhancements to P4TG: Histogram-Based RTT Monitoring in the Data Plane, in Workshop on Resilient Networks and Systems (ReNeSys), Jul. 2025, Ilmenau, Germany
@article{LiHae23,
  title   = {{P4TG: 1 Tb/s Traffic Generation for Ethernet/IP Networks}},
  author  = {Steffen Lindner and Marco Häberle and Michael Menth},
  journal = {{IEEE Access}},
  year    = 2023,
  month   = feb,
  volume  = 11,
  pages   = {17525--17535}
}

@article{IhZi25,
  title  = {{Enhancements to P4TG: Protocols, Performance, and Automation}},
  author = {Fabian Ihle and Etienne Zink and Steffen Lindner and Michael Menth},
  journal = {{KuVS Workshop on Network Softwarization (KuVS NetSoft)}},
  year   = 2025,
  month  = apr
}

@article{IhZi25_2,
  title  = {{Enhancements to P4TG: Histogram-Based RTT Monitoring in the Data Plane}},
  author = {Fabian Ihle and Etienne Zink and Michael Menth},
  journal = {{Workshop on Resilient Networks and Systems (ReNeSys)}},
  year   = 2025,
  month  = sep
}

About

This repository contains the source code for P4TG, a 1 Tb/s traffic generator for Ethernet/IP networks

Resources

Readme

License

Apache-2.0 license
Activity
Custom properties

Stars

60 stars

Watchers

5 watching

Forks

15 forks
Report repository

Releases 11

v2.7.1 Latest
Mar 24, 2026
+ 10 releases

Packages

 
 
 

Contributors

Languages