Skip to main content
AI Traffic Characterization

Introduction to the 6G Testbed: Network Emulator and AI Traffic Characterization

In short

This tutorial introduces the 6G Testbed including the Network Emulator and the framework for AI Traffic Characterization

The 6G Testbed has two parts: a Network Emulator that reproduces realistic network conditions (delay, jitter, loss, and bandwidth limits) on a single machine, and an AI Traffic Characterization framework that runs AI workloads over those conditions and logs how they behave. The video below walks through both parts and shows how they are used together.

For the full component breakdown, the profile format, and the emulator API, see the Scope page. The source code and setup instructions are in the 5G-MAG/6G-Testbed repository. The introduction video is available on the Developer Exchange section of this project's Tutorials page.

Getting started​

  1. Clone the repository: 5G-MAG/6G-Testbed.
  2. Try the network emulator on its own first, using an example profile. Pre-defined profiles are in netemu/examples/profiles.yaml, and the profiles used by the AI testbed are in aitestbed/configs/profiles.yaml.
  3. Run an AI scenario over an emulated profile using the orchestrator, then inspect the SQLite log and generated plots. To evaluate a self-hosted model without a commercial API key, use the vLLM client and a chat_vllm-style scenario.
  4. Extend it: add a scenario by subclassing BaseScenario and registering it, or add a provider by subclassing LLMClient and registering it in the orchestrator client factory (see the extension notes on the Scope page).

Applying tc impairments requires the appropriate privileges on the network interface, so run the emulator on a machine where you can configure traffic control (typically with elevated privileges), and use clear() to return the interface to normal when finished.