Abstract
We propose a novel strategy for energy-efficient dynamic computation offloading, in the context of edge-computing-aided 5G (and beyond) networks. The goal is to minimize the energy consumption of the overall system, comprising users and network elements, under constraints on the end-to-end service delay and the packet error rate performance over the wireless interface. To reduce the energy consumption, we exploit low-power sleep operation modes for the users, the access point and the edge server, shifting the edge computing paradigm from an always on to an always available architecture, capable of guaranteeing an on-demand target service quality with the minimum energy consumption. To this aim, we propose Discontinuous Mobile Edge Computing (D-MEC): an online algorithm that dynamically and optimally orchestrates the sleep mode operations and the duty cycles of the network's elements. In such a dynamic framework, end-to-end delay constraints translate into constraints on overall queueing delays, including both the communication and the computation phases of the offloading service. D-MEC hinges on stochastic Lyapunov optimization, does not require any prior knowledge on the statistics of the offloading traffic or the radio channels, and satisfies the long-term performance constraints imposed by users. Several numerical results illustrate the advantages of the proposed method.
Introduction
Mobile Edge Computing (MEC) is essential for the low-latency demands of "Beyond 5G" applications like augmented reality, autonomous driving, and industrial IoT. However, offloading computation from a mobile device to an edge server creates a massive energy footprint. While previous research often focused on saving the user's battery, it frequently ignored the energy consumed by the infrastructure—the Access Points (APs) and Edge Servers (ESs). In a typical network, these elements are "always on," consuming power even when idle.
The challenge addressed by this work is how to synchronize sleep cycles across the entire chain (User $\rightarrow$ AP $\rightarrow$ Server) to save energy without missing critical millisecond-level deadlines. We propose D-MEC (Discontinuous Mobile Edge Computing), a dynamic orchestration strategy that shifts from an "always on" to an "always available" architecture.
System Model and Methodology
The D-MEC algorithm manages radio and computational resources without needing prior knowledge of traffic patterns or radio channel statistics. It optimizes the "duty cycles" of all three entities—the User Equipment (UE), the AP, and the ES—deciding in every time slot whether they should be active or in a low-power sleep state.
Key technical components include:
- Three-Tier Queueing Model: As illustrated in the system model, data flows through three stages: the uplink queue at the device ($Q_k^l$), the computation queue at the server ($Q_k^m$), and the downlink queue at the AP ($Q_k^a$).
- Lyapunov Stochastic Optimization: This framework transforms long-term average constraints (like average delay) into a series of simple, per-slot optimization problems. It uses "virtual queues" to track performance against goals over time.
- Holistic Sleep Modes: Unlike standard models, D-MEC optimizes the duty cycles of all three entities simultaneously, accounting for transition times and control signaling ($\tau_s$).
- Adaptive Reliability Mapping: An online method translates probabilistic delay targets (e.g., "99.9% of packets must arrive within 100ms") into physical queue-length thresholds that the algorithm acts upon in real-time.
Performance Evaluation
The algorithm was evaluated using numerical simulations modeled after realistic 5G picocell environments, utilizing Rayleigh fading at 28 GHz and empirical power models for smartphones and base stations.
- Energy Efficiency: D-MEC significantly outperforms "always on" baselines and partial-sleep strategies. By optimizing the entire chain, it achieves a "globally green" state.
- Reliability: The adaptive thresholding mechanism successfully keeps the "out-of-service" probability below target thresholds (e.g., $10^{-3}$), even in fluctuating channel conditions.
- Dynamic Response: In low-traffic scenarios, the system stays in sleep mode for the majority of the time. As traffic increases, D-MEC naturally increases duty cycles to prevent queue backups, demonstrating a "fluid" response to network demand.
Key Insights and Future Work
A core insight of D-MEC is the power of the Lyapunov trade-off parameter ($V$). By tuning $V$, a network operator can precisely choose between maximum energy savings and minimum latency. Furthermore, the results show that optimizing only the user's energy is a "local" win that often leads to infrastructure waste; a holistic approach is required to reduce the overall ICT carbon footprint.
Despite its advantages, there are areas for future exploration:
- Mobility: Current focus is on a single-cell; future work must address users moving between cells and multi-cell cooperation.
- Signaling Overhead: While $\tau_s$ is accounted for, the data overhead of coordinating states in massive MIMO environments could be a bottleneck.
- Traffic Patterns: Testing against highly correlated or "heavy-tailed" traffic patterns would further validate the algorithm's robustness beyond i.i.d. assumptions.
In conclusion, D-MEC provides a scalable, computationally light solution for energy-efficient edge computing, ensuring that network infrastructure is active only when necessary while remaining always available to meet user demands.
Summary updated