[Paper Review] OpenRAN Gym: AI/ML Development, Data Collection, and Testing for O-RAN on PAWR Platforms

Leonardo Bonati, Michele Polese+3 more|arXiv (Cornell University)|Jul 25, 2022

Energy Harvesting in Wireless Networks4 citations

TL;DR

OpenRAN Gym is an open-source, O-RAN-compliant experimental toolbox that enables end-to-end AI/ML development, data collection, and testing for next-generation cellular networks on PAWR platforms. It integrates RAN statistics collection, a lightweight near-RT RIC, and xApp deployment across testbeds like Colosseum, Arena, POWDER, and COSMOS, enabling scalable, reproducible experimentation and seamless migration of AI/ML solutions from simulation to real-world deployments.

ABSTRACT

Open Radio Access Network (RAN) architectures will enable interoperability, openness and programmable data-driven control in next generation cellular networks. However, developing and testing efficient solutions that generalize across heterogeneous cellular deployments and scales, and that optimize network performance in such diverse environments is a complex task that is still largely unexplored. In this paper we present OpenRAN Gym, a unified, open, and O-RAN-compliant experimental toolbox for data collection, design, prototyping and testing of end-to-end data-driven control solutions for next generation Open RAN systems. OpenRAN Gym extends and combines into a unique solution several software frameworks for data collection of RAN statistics and RAN control, and a lightweight O-RAN near-real-time RAN Intelligent Controller (RIC) tailored to run on experimental wireless platforms. We first provide an overview of the various architectural components of OpenRAN Gym and describe how it is used to collect data and design, train and test artificial intelligence and machine learning O-RAN-compliant applications (xApps) at scale. We then describe in detail how to test the developed xApps on softwarized RANs and provide an example of two xApps developed with OpenRAN Gym that are used to control a network with 7 base stations and 42 users deployed on the Colosseum testbed. Finally, we show how solutions developed with OpenRAN Gym on Colosseum can be exported to real-world, heterogeneous wireless platforms, such as the Arena testbed and the POWDER and COSMOS platforms of the PAWR program. OpenRAN Gym and its software components are open-source and publicly-available to the research community. By guiding the readers through running experiments with OpenRAN Gym, we aim at providing a key reference for researchers and practitioners working on experimental Open RAN systems.

Motivation & Objective

• To address the lack of unified, open, and scalable platforms for developing and testing AI/ML-driven control solutions in heterogeneous Open RAN environments.
• To enable end-to-end experimentation with O-RAN-compliant xApps—from data collection and model training to deployment and runtime testing—across multiple wireless testbeds.
• To facilitate the transition of AI/ML solutions developed on emulated platforms (e.g., Colosseum) to real-world, heterogeneous testbeds such as Arena, POWDER, and COSMOS.
• To provide a standardized, open-source framework that supports interoperability, programmability, and data-driven control in next-generation cellular networks.
• To lower the barrier to entry for researchers by offering a complete, documented, and extensible toolbox for O-RAN experimentation.

Proposed method

• The framework integrates existing software stacks for RAN statistics collection and RAN control into a unified, O-RAN-compliant toolbox.
• It includes a lightweight, containerized near-RT RAN Intelligent Controller (RIC) that supports deployment and execution of xApps on experimental platforms.
• The system enables data collection from softwarized RANs via standardized E2 interfaces, supporting real-time network monitoring and telemetry.
• It supports the development, training, and testing of AI/ML-based xApps using a modular, extensible architecture that can be exported across platforms.
• The framework enables migration of experiments from the Colosseum testbed to real-world PAWR platforms (Arena, POWDER, COSMOS) via standardized LXC images and containerization.
• It provides prebuilt and build-from-scratch LXC images for the RIC and E2 termination, ensuring portability and reproducibility across testbeds.

Experimental results

Research questions

• RQ1 How can a unified, open-source platform be designed to support end-to-end AI/ML development and testing for O-RAN-compliant systems across diverse wireless testbeds?
• RQ2 What are the performance characteristics and transfer overheads of migrating LXC images and containerized RICs from Colosseum to real-world PAWR platforms such as Arena, POWDER, and COSMOS?
• RQ3 How does the performance of container instantiation and RIC build processes vary across different testbed infrastructures with varying compute capabilities?
• RQ4 To what extent can AI/ML xApps developed on emulated platforms be successfully deployed and tested on real-world, heterogeneous wireless testbeds?
• RQ5 What is the scalability and reproducibility of AI/ML workflows in O-RAN environments when using a standardized, open-source experimental framework?

Key findings

• Transfer times for LXC images from Colosseum to other testbeds ranged from approximately 1.5 minutes to nearly 6 minutes, depending on image size and platform capabilities.
• Container instantiation on Arena was significantly faster than on COSMOS and POWDER, completing in under 1 second for the SCOPE image and under 1 minute for the prebuilt RIC image, due to direct bare-metal access.
• Building the ColO-RAN RIC Docker containers from scratch took significantly longer on Arena (46 minutes) than on POWDER (21 minutes) and COSMOS (26 minutes), due to Arena’s 6-core CPU versus 24-core and 16-core servers on the other platforms.
• The prebuilt ColO-RAN LXC image enabled RIC instantiation in under 3 minutes on POWDER and COSMOS, and under 2 minutes on Arena, demonstrating efficient deployment after initial build.
• The framework successfully enabled deployment and testing of two xApps for a 7-base station, 42-user network on Colosseum, validating its capability for large-scale experimentation.
• Solutions developed on Colosseum were successfully exported and executed on the Arena, POWDER, and COSMOS testbeds, confirming cross-platform compatibility and portability.