[Paper Review] OrchestRAN: Network Automation through Orchestrated Intelligence in the Open RAN

Salvatore D’Oro, Leonardo Bonati+2 more|arXiv (Cornell University)|Jan 14, 2022

Software-Defined Networks and 5G4 citations

TL;DR

OrchestRAN is a novel orchestration framework for Open RAN that automates the deployment of data-driven AI/ML models across cloud and edge nodes based on network operators' high-level intents. It selects optimal models, execution locations, and input pipelines while meeting timing constraints and avoiding conflicts, reducing E2 interface overhead by up to 2.6× in experiments on a 7-base station testbed using Colosseum.

ABSTRACT

The next generation of cellular networks will be characterized by softwarized, open, and disaggregated architectures exposing analytics and control knobs to enable network intelligence. How to realize this vision, however, is largely an open problem. In this paper, we take a decisive step forward by presenting and prototyping OrchestRAN, a novel orchestration framework that embraces and builds upon the Open RAN paradigm to provide a practical solution to these challenges. OrchestRAN has been designed to execute in the non-real-time RAN Intelligent Controller (RIC) and allows Network Operators (NOs) to specify high-level control/inference objectives (i.e., adapt scheduling, and forecast capacity in near-real-time for a set of base stations in Downtown New York). OrchestRAN automatically computes the optimal set of data-driven algorithms and their execution location to achieve intents specified by the NOs while meeting the desired timing requirements. We show that the problem of orchestrating intelligence in Open RAN is NP-hard, and design low-complexity solutions to support real-world applications. We prototype OrchestRAN and test it at scale on Colosseum. Our experimental results on a network with 7 base stations and 42 users demonstrate that OrchestRAN is able to instantiate data-driven services on demand with minimal control overhead and latency.

Motivation & Objective

• To address the open challenge of orchestrating AI/ML intelligence in Open RAN systems with diverse timing, resource, and control requirements.
• To enable network operators to specify high-level control and inference objectives without low-level deployment decisions.
• To minimize control overhead and latency by intelligently placing models at the optimal network location—cloud or edge—based on timing and resource constraints.
• To prevent conflicts between co-located models controlling the same RAN parameters.
• To provide a practical, scalable, and O-RAN-compliant solution for real-time and near-real-time network automation.

Proposed method

• OrchestRAN operates within the non-Real-time RIC and uses a model selection and placement algorithm to map high-level operator intents to specific data-driven models.
• It formulates the orchestration problem as a constrained optimization task, proven to be NP-hard, and designs low-complexity heuristics for real-time feasibility.
• The framework leverages O-RAN’s open E2 interface to collect inputs and send control commands, ensuring interoperability with standard RAN components.
• It supports both xApps (executed at the RIC) and dApps (executed at DUs), enabling distributed execution of AI/ML workloads.
• The system dynamically instantiates and manages models based on timing requirements, resource availability, and input data accessibility.
• It includes conflict detection and mitigation logic to ensure that multiple models do not interfere when controlling shared RAN parameters.

Experimental results

Research questions

• RQ1 How can network operators specify high-level control and inference objectives in Open RAN without managing low-level deployment details?
• RQ2 What is the optimal placement of data-driven models (cloud vs. edge) to meet timing, resource, and input availability constraints?
• RQ3 How can conflicts between multiple AI/ML models controlling the same RAN parameters be detected and resolved?
• RQ4 What is the impact of edge-based model execution on control loop latency and interface overhead in the O-RAN E2 interface?
• RQ5 Can a practical orchestration framework achieve low-complexity, scalable, and conflict-free deployment of AI/ML services in real-world Open RAN deployments?

Key findings

• OrchestRAN successfully instantiates data-driven services on demand with minimal control overhead and latency in a 7-base station, 42-UE testbed on Colosseum.
• The framework reduces E2 interface traffic by up to 2.6× when deploying models at the DU (edge) compared to deploying only at the RIC.
• Approximately 25% of models are offloaded to DUs in configurations with edge compute, while 75% are executed at the DU, reducing RIC load and E2 signaling.
• Payload data accounts for about 40% of E2 traffic, with 60% being control and management overhead, which is significantly reduced through intelligent orchestration.
• OrchestRAN enables seamless dynamic activation of xApps and dApps on the same DU without service interruption, demonstrating real-time adaptability.
• In network slicing scenarios, M4 (with slicing control) achieves up to 10% higher throughput for slice 0 and 2× more transmitted packets for slice 1 than M3, showing superior performance with integrated policy control.