Optimizing energy performance, resources, and the user experience through the orchestration of network slicing architecture
This paper proposes a novel End-to-End Energy-Efficient Orchestration (E2E-EEO) architecture that leverages AI-native orchestration and joint resource optimization to resolve the critical trade-off between energy sustainability and Quality of Service in 5G network slicing.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The world is becoming increasingly connected, with billions of devices relying on wireless networks to send data, stream video, and control machinery. As these networks evolve to handle more traffic and faster speeds, they consume vast amounts of electricity, creating a significant environmental footprint. To manage this complexity, engineers use a concept called network slicing. Imagine a single physical road that can be instantly divided into separate, invisible lanes, each designed for a specific type of traffic: one lane for high-speed video, another for critical industrial controls, and a third for simple text messages. This approach allows different services to share the same infrastructure without interfering with one another. However, a paradox has emerged. While these separate lanes make the network more flexible, the traditional way of managing them often wastes energy. To guarantee that every lane is always ready, the system keeps resources reserved even when no one is using them, much like leaving every light on in a building just in case someone walks in. This static approach creates a difficult trade-off: saving energy by turning off unused parts of the network risks slowing down the services people rely on, while keeping everything running at full speed guarantees performance but burns unnecessary power.
Researchers at the Universidade de São Paulo have proposed a new way to solve this problem, moving away from rigid, separate management systems toward a unified, intelligent approach. In a preprint posted on October 12th, 2026, Marcia Cristina Machado and Tereza Cristina M. B. Carvalho describe a framework called End-to-End Energy-Efficient Orchestration. Their work addresses the core challenge of balancing energy savings with the need for high-quality service across the entire network, from the cell towers that connect to your phone to the central data centers that process information. The authors argue that current methods are too fragmented, often optimizing one part of the network in isolation, which can accidentally cause problems or waste energy in another part. Instead, they suggest a system that looks at the whole network as a single, connected entity, using artificial intelligence to make dynamic decisions in real time.
The proposed architecture relies on a hierarchical structure that brings together different layers of control. At the top sits a "Hyper-Orchestrator," a central brain that coordinates the entire network, while specialized controllers manage specific sections like the radio access network, the transport links, and the core data centers. This setup allows the system to make global decisions that consider the impact of every action across all domains. For instance, if the system decides to put a base station into a low-power sleep mode to save energy, the central controller ensures that this action does not overload the transport network or degrade the experience for users on a critical slice. This holistic view prevents the common issue where saving energy in one area simply shifts the burden and increases consumption elsewhere.
A key innovation in this framework is the use of artificial intelligence, specifically machine learning, to manage the network without constant human intervention. The system learns from real-time data to predict traffic patterns and resource needs before they happen. Instead of keeping resources fixed and ready for the busiest possible moment, the AI dynamically scales them up or down based on actual demand. It can activate or deactivate entire network slices depending on whether they are being used, and it can even place network functions, such as virtual firewalls or routers, on servers where they will run most efficiently. The researchers note that this intelligent management could reduce manual tasks by up to 70 percent and improve resource utilization by between 35 and 45 percent compared to older, static methods.
To ensure that these energy-saving measures do not hurt the user, the architecture incorporates a direct line of communication between the applications running on the network and the network itself. This means that if a user is running a cloud virtual reality application that requires ultra-low latency, the network knows exactly what that application needs and prioritizes resources accordingly. Conversely, for less demanding traffic, the system can apply more aggressive energy-saving policies. This application-aware approach allows the network to maintain a high quality of experience for critical services while still conserving power. The researchers also emphasize that security must be built into this design from the start, ensuring that the virtual separation between different slices remains robust against threats like data leakage or attacks, even as the network changes shape to save energy.
The study validates these ideas by referencing quantitative results from existing literature. The authors point out that previous studies have shown that joint optimization across the entire network can reduce total energy consumption significantly compared to methods that only focus on the radio access part. For example, simulations of dynamic slice activation have demonstrated energy efficiency improvements of 11 to 14 percent while maintaining service quality at similar levels. Other research cited in the paper indicates that application-aware orchestration can improve the quality of experience for web traffic by up to 50 percent. These findings, drawn from the literature, suggest that the principles of the proposed architecture are not just a theoretical concept but a practical solution that has been shown to deliver measurable improvements in real-world scenarios.
The researchers conclude that while network slicing is essential for the future of 5G and beyond, its current implementation often leads to inefficiencies that can be corrected with a smarter, more integrated approach. Their work highlights that the problem is not the existence of separate network lanes, but the lack of an intelligent layer that can manage them dynamically with energy as a primary goal. By combining a hierarchical control structure with artificial intelligence and direct application feedback, the proposed framework offers a path toward a truly sustainable network. The authors also look ahead, suggesting that future developments could integrate even more advanced technologies, such as large language models to translate human intent into network commands, or quantum computing to solve the complex mathematical problems involved in optimizing such vast systems. Ultimately, this research provides a clear blueprint for how the digital infrastructure of the future can be both powerful and environmentally responsible.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.