Joint User Scheduling and Computing Resource Allocation Optimization in Asynchronous Mobile Edge Computing Networks

Zhaohui Yang, Yihan Cang, Ming Chen 0001, Yijin Pan, Zhaohui Yang 0001, Ye Hu, Haijian Sun, Mingzhe ChenIEEE Trans. Communications 2024 · DOI

TL;DR

This paper introduces a novel asynchronous mobile edge computing (MEC) framework that optimizes both user task offloading schedules and computation resource allocation to minimize energy consumption. By leveraging wireless power transfer and a novel decomposition method, it achieves significant energy savings compared to traditional synchronous MEC systems.

What Problem Does This Paper Address?

Mobile edge computing (MEC) offers powerful computing capabilities to edge devices by offloading tasks to nearby MEC servers. However, existing MEC systems often assume synchronous task arrivals and processing, which is inefficient and impractical given the dynamic nature of mobile devices. Furthermore, edge servers have limited energy and computational resources, leading to challenges in providing sufficient processing power and managing task scheduling effectively. This paper addresses the problem of optimizing both when devices offload their tasks (scheduling) and how much computational resource is allocated to them, all within an asynchronous framework that considers energy harvesting and limited server capacity. The goal is to minimize the total energy consumed by mobile computing tasks while meeting latency requirements.

Methods

The core of this paper lies in its novel asynchronous MEC framework and the optimization approach.

  • Asynchronous Framework: Unlike synchronous systems, this framework allows tasks to arrive and be processed at the MEC server asynchronously. This is illustrated by a flowchart showing devices harvesting energy, offloading tasks at different times, and the server processing them sequentially as they arrive.
  • Joint Optimization: The problem of joint user scheduling, time allocation, and computation resource allocation is formulated as a mixed-integer non-linear programming (MINLP) problem.
  • General Benders Decomposition (GBD): To tackle the complexity of the MINLP, the authors employ a GBD method. This decomposes the problem into:
  • Primal Problem: This subproblem focuses on allocating computation resource and time slots for a fixed user schedule. The authors derive closed-form solutions for optimal computation resource allocation and propose a low-complexity algorithm (Algorithm 1) that leverages properties of the optimal solution to significantly reduce computational complexity compared to interior-point methods.
  • Master Problem: This subproblem optimizes the user scheduling by iteratively adding optimality or feasibility cuts based on the solutions from the primal problem. This master problem is a mixed-integer linear programming (MILP) problem, which can be efficiently solved.
  • Wireless Power Transfer (WPT): The framework incorporates WPT, allowing edge devices to harvest energy from ambient RF signals to power their task offloading, thus enhancing energy efficiency.

Datasets

Not discussed in the paper. The paper relies on simulated data for evaluation.

Key Insights and Ideas

  • Asynchronicity is Key: The paper strongly advocates for asynchronous processing in MEC, demonstrating its significant energy efficiency benefits over synchronous approaches.
  • Decomposition for Tractability: The GBD approach effectively breaks down a complex MINLP into manageable primal and master problems, enabling efficient joint optimization.
  • Properties of Optimal Resource Allocation: The authors derive crucial properties of the optimal computation resource allocation, such as the "transition point" where frequency allocation changes. This insight is leveraged to develop a low-complexity algorithm.
  • WPT as an Enabler: Integrating WPT is shown to be a critical component for achieving energy-efficient asynchronous MEC.

Key Results

  • Significant Energy Reduction: The proposed asynchronous computing framework achieves an impressive 87.87% energy consumption reduction compared to conventional synchronous computing counterparts.
  • Computational Complexity Reduction: The proposed optimal computation resource allocation algorithm (Algorithm 1) offers a 100-times reduction in computational complexity compared to conventional interior-point methods.
  • Near-Optimal Performance: Simulation results show that the proposed algorithm achieves results very close to exhaustive search, indicating near-optimal performance.
  • Performance Gains over Benchmarks: The proposed scheme outperforms JSORA (30.88% energy reduction), random scheduling (19.51% energy reduction), and JCCRM-Sync (87.87% energy reduction).

Limitations and Future Work

  • Generalization to Heterogeneous Deadlines: The authors suggest extending the asynchronous framework to scenarios with heterogeneous task deadlines.
  • Online Algorithm Design: Investigating online algorithm design to accommodate new devices joining the network dynamically is proposed as a future direction.
  • Multi-Server Systems: The paper briefly mentions that for multi-server MEC systems, new indicator variables could be introduced, and the problem could be solved using advanced MINLP methods or machine learning. The detailed transmission protocol and algorithm procedure for such systems are left for future work.

Summary updated

BibTeX
@article{yang2024joint,
  author = {Zhaohui Yang and Yihan Cang and Ming Chen and Yijin Pan and Zhaohui Yang and Ye Hu and Haijian Sun and Mingzhe Chen},
  title = {{Joint User Scheduling and Computing Resource Allocation Optimization in Asynchronous Mobile Edge Computing Networks}},
  journal = {IEEE Trans. Communications},
  year = {2024},
  doi = {10.1109/tcomm.2024.3358237},
}
Figure