Fair Resource Allocation in Virtualized O-RAN Platforms

TL;DR

This work addresses the rising energy costs of virtualized O-RAN (vRAN) platforms by proposing a horizon-fair, online control framework that jointly optimizes compute-load distribution across heterogeneous O-Cloud processing units and near-real-time TB size thresholds for user transmissions. It introduces a saddle-point reformulation with proxy functions $\Psi_t(\bm \theta,\bm \phi,\bm x)$ and dual variables in bounded sets $\Theta$ and $\Phi$, enabling optimistic Follow-The-Regularized-Leader (OFTRL) updates on both the primal and dual spaces. The methods yield sublinear horizon fairness regret $\bm{\mathcal{R}}_T(F_{\alpha},F_{\beta})$ in adversarial settings and approach constant regret $\mathcal{O}(1)$ under perfect predictions, with closed-form, lightweight updates suitable for non-RT and near-RT RICs. Evaluation on trace-driven simulations and an O-RAN testbed demonstrates energy savings up to about $72\%$ with reasonable latency tradeoffs, validating the practicality and impact of horizon-fair resource allocation in O-Cloud vRANs.

Abstract

O-RAN systems and their deployment in virtualized general-purpose computing platforms (O-Cloud) constitute a paradigm shift expected to bring unprecedented performance gains. However, these architectures raise new implementation challenges and threaten to worsen the already-high energy consumption of mobile networks. This paper presents first a series of experiments which assess the O-Cloud's energy costs and their dependency on the servers' hardware, capacity and data traffic properties which, typically, change over time. Next, it proposes a compute policy for assigning the base station data loads to O-Cloud servers in an energy-efficient fashion; and a radio policy that determines at near-real-time the minimum transmission block size for each user so as to avoid unnecessary energy costs. The policies balance energy savings with performance, and ensure that both of them are dispersed fairly across the servers and users, respectively. To cater for the unknown and time-varying parameters affecting the policies, we develop a novel online learning framework with fairness guarantees that apply to the entire operation horizon of the system (long-term fairness). The policies are evaluated using trace-driven simulations and are fully implemented in an O-RAN compatible system where we measure the energy costs and throughput in realistic scenarios.

Figures (13)

• Figure 1: Cell load dynamics (msec granularity) over a few seconds, collected from an operational RAN in Frankfurt, Germany, May 2023.
• Figure 2: Processing latency (left) and energy consumption (right) to process one TB under different SNRs; measured on an Intel Xeon CPU core and an NVIDIA V100 GPU.
• Figure 3: (a): A non-RT controller at the Service & Management Orchestration (SMO) framework devises the load assignment policy every $\sim \!1\!-\!10$ seconds and sends it to the vBSs via the A1 interface. (b): Timing diagram of assignment implementation and learning policy.
• Figure 4: (a): The near-RT controller decides the TB threshold (minTB) policy for each user at each slot; and the TBs are processed at a HA-equiped server. (b): Timing diagram of the learning algorithm for the minTB policy.
• Figure 5: Schematic of the experimental platform, including the RICs and interfaces of the use cases.

Theorems & Definitions (7)

• lemma 1
• lemma 2
• Proposition 1
• Proposition 2
• theorem 1
• theorem 2
• lemma 3