💻 computer science

Latency-Aware and Energy-Efficient Workload Distribution in Hierarchical Vehicular Fog Computing

This paper proposes a latency-aware and energy-efficient workload distribution framework for hierarchical vehicular fog computing that utilizes a Modified Particle Swarm Optimization (MPSO) algorithm to optimize task offloading and resource allocation, demonstrating significant improvements in latency, energy consumption, and execution time compared to conventional methods.

Original authors: Salman Khan, Woong-Kee Loh, Muhammad Babar

Published 2026-10-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Salman Khan, Woong-Kee Loh, Muhammad Babar

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

Modern cars are becoming less like simple machines and more like rolling computers. They are packed with sensors that constantly scan the road, detect obstacles, and calculate the safest path forward. To make these decisions instantly, the car's onboard computer must process a massive stream of data. However, the processors inside a vehicle have limits. When the amount of information grows too large or the need for speed becomes critical, the car's own computer can become overwhelmed, leading to dangerous delays. Sending all this data to a distant cloud server for processing is not a solution either, because the time it takes for the signal to travel back and forth is too long for split-second driving decisions. This creates a bottleneck where the need for instant reaction clashes with the physical limits of computing power and distance.

To solve this, researchers have turned to a concept called fog computing. Imagine the cloud as a massive, centralized library far away, and the car as a reader who needs a book right now. Fog computing places a small, local branch of that library right next to the reader, perhaps at a street corner or a traffic light. In this setup, the "fog" consists of computing resources located close to the vehicles, such as roadside units or other nearby vehicles, which can handle urgent tasks immediately without waiting for a distant server. This approach promises to keep self-driving cars safe and responsive, but it introduces a new challenge: how to decide which nearby computer should handle which task. If the system chooses poorly, it wastes energy or causes a delay that could be fatal.

In a recent study, a team of researchers from Gachon University and the Middle East College proposed a new way to manage this complex distribution of work. They focused on a three-layer system: the vehicles themselves at the bottom, the local fog nodes in the middle, and the massive cloud data centers at the top. Their goal was to create a smart system that could automatically decide, in real-time, whether a car should process a task itself, send it to a nearby fog node, or forward it to the cloud. The researchers developed a specific method, which they call a modified particle swarm optimization algorithm, to act as the brain behind these decisions. This method mimics how a flock of birds might search for food; instead of every bird flying randomly, they share information about the best spots found so far to guide the whole group toward the most efficient solution. In this case, the "birds" are virtual search agents looking for the perfect balance between speed and energy use.

The researchers tested their idea using a sophisticated simulation tool designed specifically for fog computing environments. They created a virtual world with 1,000 fog nodes and 250 vehicles, simulating a busy, real-world traffic scenario where thousands of tasks needed processing. They compared their new method against several other existing strategies used by computers to solve similar problems. The results showed that their approach was significantly more effective. In these simulations, the new method reduced the average time it took for a task to be completed by nearly 24 percent compared to standard methods. It also cut the energy used by the system by over 22 percent, a crucial factor since fog devices often run on limited power. Furthermore, the system reached a decision much faster, taking about 32 percent less time to figure out the best plan than the older techniques.

Beyond just speed and power, the study found that the new method was better at keeping the system running smoothly. It improved the rate at which tasks were successfully completed by about 4.5 percent, meaning fewer requests were dropped or failed. It also managed the available computing resources more efficiently, utilizing them nearly 10 percent better than the alternatives. This means the system could handle more cars and more data without getting bogged down. The researchers noted that while other methods sometimes struggled to find the best option quickly, their approach consistently found a balanced solution that minimized delays and energy use simultaneously. The study did not test this on real cars on a real highway; instead, it relied entirely on these detailed computer simulations to prove the concept works under controlled conditions.

The significance of this work lies in its ability to handle the chaotic and fast-changing nature of traffic. Unlike a static office network, a road network is constantly shifting as vehicles move in and out of range. The researchers' method is designed to adapt to these changes instantly, ensuring that a car approaching a busy intersection gets the computing help it needs exactly when it needs it. By proving that this specific algorithm can outperform existing techniques in a simulated environment, the study offers a promising blueprint for the future of intelligent transportation. It suggests that with the right software, the fog layer can become a reliable, energy-efficient partner to our vehicles, keeping the flow of information as smooth as the flow of traffic itself.

Problem Statement

The rapid evolution of the automotive sector and the proliferation of Intelligent Transportation Systems (ITS) have created a demand for computational architectures capable of ultra-low latency and high processing efficiency. While Cloud Computing offers massive storage and processing power, its geographical distance from end-user devices results in significant multi-hop communication delays, making it unsuitable for delay-sensitive vehicular applications. Conversely, while Fog Computing brings resources closer to the edge, Fog nodes are inherently resource-constrained regarding computation, storage, and energy.

The core challenge addressed is the efficient management of these constrained resources in a dynamic, mobile environment. Specifically, the paper identifies the need for a mechanism to handle the massive influx of data generated by connected vehicles (projected to reach 73 ZB by 2025) without incurring unacceptable latency or excessive energy consumption. Existing solutions often fail to adequately balance the trade-offs between latency, energy dissipation, and execution time in real-time vehicular scenarios.

Methodology

To address these challenges, the authors propose a hierarchical tri-tier architecture consisting of:

IoT Layer (Bottom): Mobile vehicles equipped with sensors (LiDAR, radar, proximity) acting as data acquisition nodes.
Fog Layer (Middle): Intermediate computing nodes (e.g., Roadside Units) that process requests locally to minimize latency.
Cloud Layer (Top): Centralized data centers for long-term storage and complex, non-real-time analytics.

The core of the proposed solution is a Modified Particle Swarm Optimization (MPSO) algorithm designed for dynamic resource provisioning and task offloading. The methodology operates as follows:

Workflow: Vehicle sensors generate tasks which are buffered and forwarded to a Fog Broker/Dispatcher. The broker evaluates the current load and resource availability of Fog nodes.
Decision Logic: If a Fog node is overloaded, tasks are offloaded to the Cloud. If the Fog node has capacity, the system employs a "Best-Fit" search within the neighborhood. If a suitable node is not immediately found, the MPSO algorithm is triggered.
MPSO Mechanism: The algorithm utilizes a local neighborhood search to identify a personal best ( $P_{best}$ ) and a global best ( $G_{best}$ ) resource allocation. It iteratively refines these positions to find a Pareto-optimal solution that balances conflicting objectives.
Optimization Model: The problem is formulated as a multi-objective optimization task aiming to minimize:
- Total Latency ( $L(x)$ ): Including transmission, queuing, propagation, and computation delays.
- Total Energy Consumption ( $E(x)$ ): Accounting for CPU usage, communication, memory, and networking overheads.
- Computational Cost ( $C(x)$ ).
  The final solution is selected using normalized objective values weighted by preference coefficients ( $\alpha, \beta, \gamma$ ).

Key Contributions

The paper enumerates three primary contributions:

Novel Architecture: A proposed tri-layered (Cloud-Fog-IoT) framework specifically designed to support robust, real-time vehicular applications.
Advanced Offloading Strategy: The introduction of an MPSO-based algorithm tailored for latency-sensitive environments. This approach aims to accelerate convergence, reduce latency and energy costs, and maximize task success rates compared to standard heuristics.
Empirical Validation: A comprehensive evaluation using the iFogSim simulation toolkit, comparing the proposed MPSO against conventional Particle Swarm Optimization (PSO), Genetic Algorithms (GA), Ant Colony Optimization (ACO), Artificial Bee Colony (ABC), and Simulated Annealing.

Experimental Results

The study was simulated with 1,000 Fog nodes and 250 vehicles. The results indicate that the proposed MPSO framework outperforms existing meta-heuristic algorithms across all measured metrics:

Latency: Reduced average latency by 23.8% compared to conventional PSO.
Energy Consumption: Decreased energy usage by 22.1%.
Execution Time: Reduced the time required to make offloading decisions by 32.4% (9.2s for MPSO vs. 13.6s for PSO).
Convergence: Achieved convergence in 34.9% fewer iterations (41 iterations vs. 63 for PSO).
Resource Utilization: Improved by 9.7%, reaching 91.6%.
Task Success Rate: Improved by 4.5%, reaching 97.3%.

Significance and Claims

The authors claim that the proposed MPSO framework provides a superior solution for large-scale vehicular fog-cloud computing by offering faster convergence, lower delay, and reduced energy usage. The study asserts that the integration of local neighborhood search with global best refinement allows the system to handle the dynamic nature of vehicular networks more effectively than static or single-objective approaches.

The paper concludes that the proposed methodology successfully mitigates the computational overhead across Fog nodes while ensuring efficient resource utilization. It positions the work as a step toward enabling reliable, real-time task processing for autonomous vehicles, addressing the critical constraints of energy and latency that currently hinder the deployment of advanced ITS. The authors note that while the algorithm shows linear efficiency in best-case scenarios, it operates within a quadratic regime under typical and adverse conditions, reflecting the inherent complexity of iterative resource searching in dynamic environments.

Problem Statement

Methodology

Key Contributions

Experimental Results

Significance and Claims

More like this