Joint Node Selection and Resource Allocation for Task Offloading in Scalable Vehicle-Assisted Multi-Access Edge Computing

Xuan-Qui Pham, Tien-Dung Nguyen, Vandung Nguyen, Eui-Nam HuhSymmetry 2019 · DOI

TL;DR

This paper tackles the challenge of resource limitations in Multi-Access Edge Computing (MEC) for Internet of Things (IoT) devices by proposing a novel "Scalable Vehicle-Assisted MEC" (SVMEC) paradigm. SVMEC integrates MEC with remote cloud and mobile volunteer nodes (MVNs, like buses) to dynamically offload tasks, minimizing total computation overhead (task completion time and monetary cost).

What Problem Does This Paper Address?

The proliferation of IoT devices is leading to an explosion of data and computationally intensive applications, such as in healthcare and transportation. While MEC offers low-latency processing closer to users, its resources are often limited, leading to bottlenecks and delays when faced with a surge of requests. Traditional solutions like mobile cloud computing (MCC) suffer from high latency due to the distance to the cloud. This paper addresses the need for a more scalable and efficient approach to task offloading that leverages both fixed edge resources (MEC), powerful but costly cloud resources, and underutilized mobile resources (vehicles). The core problem is how to intelligently decide where to offload tasks and how much resource to allocate to each, considering both time and cost constraints.

Methods

The authors propose a Scalable Vehicle-Assisted MEC (SVMEC) paradigm. The core innovation lies in a two-stage optimization approach to solve the Joint Node Selection and Resource Allocation (JNSRA) problem, formulated as a Mixed Integer Nonlinear Programming (MINLP) problem.

  1. Computing Resource Allocation: Given a fixed task offloading decision (i.e., which node will handle which task), the problem of allocating computational resources to those tasks is convex. The authors solve this using Karush-Kuhn-Tucker (KKT) conditions to find the optimal resource allocation.
  2. Node Selection: Based on the optimal resource allocation from the first stage, the problem is transformed into a 0-1 integer programming problem for selecting the computing nodes. This is then solved using a branch-and-bound algorithm.

This decomposition allows for an efficient solution to the complex JNSRA problem. The SVMEC model specifically utilizes buses as Mobile Volunteer Nodes (MVNs) due to their predictable trajectories and public nature, which simplifies privacy concerns and incentive mechanisms.

Datasets

The paper uses the Seattle bus trace dataset [43] to simulate the movement of buses, which serve as MVNs. This dataset provides real-world mobility information for approximately 1200 city buses over several weeks. The authors sampled a specific 10-minute window for their experiments.

Key Insights and Ideas

  • Vehicles as a Scalable Resource: The paper highlights the significant potential of leveraging underutilized resources in vehicles (like buses) to enhance the scalability of edge computing. This moves beyond traditional fixed MEC servers and remote clouds.
  • Joint Optimization is Crucial: Simply offloading tasks randomly or to the closest available node is suboptimal. The paper demonstrates that jointly optimizing where to offload and how much resource to allocate is key to minimizing overall computation overhead.
  • Trade-off Between Time and Cost: The SVMEC paradigm explicitly models the trade-off between task completion time and the monetary cost of using cloud or MVN resources. The weighted parameters allow the MEC provider to prioritize one over the other.
  • MVN Suitability Varies: The paper shows that MVNs are more suitable for tasks with smaller input data sizes due to transmission time considerations. As input data size increases, tasks are more likely to be offloaded to MEC or the remote cloud.
  • Bus Mobility is Advantageous: The predictable nature of bus routes makes them ideal MVNs, allowing for better estimation of their availability and transmission rates.

Key Results

The proposed SVMEC scheme consistently outperforms other baseline schemes (MEC only, MEC + Cloud, and Random Offloading in MEC + Cloud + MVNs) in terms of total computation overhead.

  • Reduced Overhead: The SVMEC scheme achieves the lowest total computation overhead across various scenarios, including varying numbers of tasks, input data sizes, and computation intensities.
  • Scalability: The system demonstrates good scalability, with total computation overhead increasing less dramatically than other schemes as the number of tasks grows.
  • Task Distribution: The paper visualizes how tasks are distributed among MEC, remote cloud, and MVNs. Notably, MVNs are utilized more when the number of tasks is moderate, and their utilization decreases when tasks become very numerous or have large data sizes.
  • Parameter Sensitivity: The weighted parameters for completion time and monetary cost significantly influence the achieved overhead, highlighting the importance of tuning these based on the MEC provider's priorities.

Limitations and Future Work

  • QoS Requirements: The current model primarily focuses on minimizing total computation overhead. Future work could extend the model to consider various Quality of Service (QoS) requirements for task requests.
  • MVN Diversity: While the paper focuses on buses, it acknowledges that MVNs could include other vehicle types (cars, trucks). Further investigation is needed to handle the diverse mobility trajectories of these vehicles.
  • Incentive Mechanism: The paper assumes a basic incentive mechanism for MVNs. More sophisticated incentive schemes could be explored to further encourage resource sharing.

Summary updated

BibTeX
@article{pham2019joint,
  author = {Xuan-Qui Pham and Tien-Dung Nguyen and Vandung Nguyen and Eui-Nam Huh},
  title = {{Joint Node Selection and Resource Allocation for Task Offloading in Scalable Vehicle-Assisted Multi-Access Edge Computing}},
  journal = {Symmetry},
  year = {2019},
  doi = {10.3390/sym11010058},
}
Figure