[Paper] Real-Time Service Subscription and Adaptive Offloading Control in Vehicular Edge Computing

Source: arXiv - 2512.14002v1

Overview

The paper tackles a core bottleneck in Vehicular Edge Computing (VEC): how to decide, in real time, which vehicle‑generated tasks should be processed locally and which should be offloaded to nearby Roadside Units (RSUs) while respecting tight deadlines and limited bandwidth. By formulating the problem as a deadline‑constrained task‑offloading and resource‑allocation optimization (DOAP) and introducing a new approximation algorithm, the authors demonstrate measurable gains for latency‑sensitive services such as on‑board object detection.

Key Contributions

• Formal DOAP model that captures both bandwidth and compute constraints, vehicle utility, and hard task deadlines.
• SARound algorithm: a linear‑program‑based rounding + local‑ratio technique that improves the theoretical approximation guarantee from 1/6 to 1/4.
• Online subscription & offloading framework that dynamically adapts to fluctuating wireless conditions and short‑deadline tasks.
• VecSim simulator built on OMNeT++ and Simu5G, integrating the full task life‑cycle (generation, subscription, offloading, execution, and result collection).
• Extensive evaluation using real taxi traces and profiled object‑detection workloads, showing consistent utility improvements over state‑of‑the‑art baselines with low runtime overhead.

Methodology

1. Problem Formulation – The authors model each vehicle’s request as a tuple (data size, compute load, deadline, utility). The goal is to maximize the sum of utilities of tasks that meet their deadlines, subject to:
   • Bandwidth caps on the wireless link between vehicle and RSU.
   • CPU core limits on the RSU’s edge server.
2. Approximation via LP Rounding – They first relax the integer program to a linear program, solve it efficiently, and then apply a randomized rounding scheme that respects the resource caps. A local‑ratio step refines the solution, guaranteeing that at least 25 % of the optimal utility is achieved (the 1/4 approximation).
3. Online Control Loop – As vehicles move, the framework continuously:
   • Subscribes vehicles to a service class based on their current utility‑to‑deadline ratio.
   • Predicts short‑term channel quality using recent SNR measurements.
   • Decides offload vs. local execution using the SARound solution as a decision oracle.
4. Simulation Environment – VecSim reproduces realistic vehicular mobility (NYC taxi trace), 5G NR channel models (via Simu5G), and edge server scheduling. The object‑detection task is profiled on an NVIDIA Jetson‑class device to obtain realistic compute/communication footprints.

Results & Findings

Key takeaways

• The improved approximation ratio translates into tangible utility gains even under highly variable bandwidth.
• The online framework reacts within milliseconds to sudden drops in SNR, preventing deadline violations.
• SARound’s computational overhead remains negligible, making it feasible for on‑board controllers or RSU edge managers.

Practical Implications

• Edge Platform Vendors can embed SARound as a lightweight scheduler in RSU software stacks, offering a provable performance guarantee without heavyweight optimization solvers.
• Automotive Developers building ADAS or infotainment services can rely on the subscription model to prioritize safety‑critical tasks (e.g., pedestrian detection) while offloading less urgent analytics.
• Network Operators gain a tool to better allocate 5G slice resources for vehicular services, potentially reducing over‑provisioning by exploiting the algorithm’s efficient resource packing.
• Simulation & Testing – VecSim provides a ready‑to‑use benchmark suite for researchers and engineers to evaluate new VEC protocols under realistic mobility and radio conditions.

Limitations & Future Work

• Static RSU Placement: The study assumes fixed RSU locations; dynamic placement or mobile edge nodes (e.g., drones) are not considered.
• Single‑Task Per Vehicle: The model handles one active task per vehicle; extending to concurrent multi‑task scenarios could affect utility trade‑offs.
• Channel Prediction Simplicity: The online controller uses short‑term SNR averaging; more sophisticated machine‑learning predictors might further reduce miss rates.
• Scalability to City‑Scale Deployments: While simulations use a large trace, real‑world deployments with thousands of RSUs may require hierarchical coordination, an avenue the authors suggest for future exploration.

Authors

• Chuanchao Gao
• Arvind Easwaran

Paper Information

• arXiv ID: 2512.14002v1
• Categories: cs.DC, cs.DM
• Published: December 16, 2025
• PDF: Download PDF