MOSAIC: Multi Operator Sharing and Allocation of RAN Resources in Open-RAN

Abstract

This paper addresses the challenge of sharing Radio Access Network (RAN) resources among multiple operators in a software-based Open-RAN (O-RAN) and 5G Core environment. Using the Multi-Operator Core Network (MOCN) architecture, traffic from different Core networks is served by a shared O-RAN gNodeB (gNB), using a unified scheduler and a common pool of physical radio resources. We introduce MOSAIC, a framework that leverages the Core-to-RAN Shaping Rate (CRSR) of each flow to dynamically manage shared RAN resources without modifying the gNB's internal scheduling policies. Under MOSAIC, Core networks regulate their respective traffic rates to ensure fair resource allocation while satisfying delay requirements of user flows. We design and study two decentralized design variants for CRSR control: (a) MOSAIC-ID, a lightweight adaptive increase-decrease scheme requiring no state feedback from the gNB; (b) MOSAIC-DML, a machine learning (ML)-based approach where CRSR is computed at each Core using estimated or signaled flow-specific gNB state information. We evaluate the two MOSAIC algorithms using a simulation with up to five flows per Core. We further implement and compare algorithms on a 5G testbed compliant with 3GPP using OpenAirInterface (OAI) [14] gNB and L25GC+ [18] Core, demonstrating the value of using an ML-based CRSR selection. We quantify the tradeoff between packet-delay violations and CRSR across uncongested, moderately congested, and highly congested traffic scenarios.