Joint Latency-Energy Minimization for Fog-Assisted Wireless IoT Networks

Farshad Shams, Vincenzo Lottici, Zhi Tian, Filippo Giannetti

TL;DR

This paper tackles the challenge of simultaneously minimizing latency and energy consumption in fog-assisted IoT networks. It introduces a novel resource allocation framework that uses a practical power consumption model and a cooperative game theory approach (Nash Bargaining Solution) to find an optimal trade-off, leading to more efficient and fair resource distribution for IoT devices.

What Problem Does This Paper Address?

The proliferation of Internet of Things (IoT) devices has led to an explosion of data and a growing demand for efficient processing. Traditional cloud computing can introduce significant latency due to data transmission distances. Fog computing, which brings computation closer to the data source (e.g., at the base station), offers a solution. However, optimizing resource allocation in these fog-assisted networks is complex. Specifically, IoT devices face a fundamental trade-off: faster task execution (lower latency) often requires more energy, while conserving energy can lead to higher latency. Existing research often simplifies power consumption models or focuses on optimizing only one objective (latency or energy) or a simple weighted sum, which doesn't fully capture the intricate relationship between these two critical metrics. Furthermore, some approaches can lead to unfair resource allocation among devices. This paper aims to address these limitations by developing a method that jointly minimizes both latency and energy consumption, considers realistic power models, and ensures fair resource distribution.

Methods

The paper proposes a multi-faceted approach:

  • Bi-Objective Optimization: Instead of optimizing for a single goal or a simple weighted sum, the authors formulate the problem as minimizing both service latency ($T_k$) and energy consumption ($E_k$) for each device $k$.
  • Pareto Boundary Derivation: To understand the trade-off between latency and energy, they use the Tchebyshev method. This method helps derive the Pareto boundary, which represents the set of optimal solutions where no improvement can be made in one objective without sacrificing the other.
  • Practical Power Consumption Models: A key contribution is the use of realistic models for the power consumption of IoT devices. Unlike previous works that often assume fixed circuit power, this paper accounts for how power consumption varies with transmit power and data rates, making the model more accurate.
  • Convexification: The initial optimization problem is non-convex. The authors employ a convexification technique to transform it into a solvable convex problem, ensuring that the optimization process can find a globally optimal solution.
  • Cooperative Game Theory (Nash Bargaining Solution - NBS): To address the competition among devices for limited fog resources and ensure fairness, the problem is framed as a cooperative game. The Nash Bargaining Solution (NBS) is used to find a unique and optimal agreement among devices, promoting proportional fairness.
  • Block Coordinate Descent Algorithm: To efficiently compute the cooperative Nash Equilibrium (NE), a block coordinate descent algorithm is developed. This iterative algorithm breaks down the complex optimization problem into smaller, manageable convex sub-problems that can be solved efficiently.

Datasets

Not discussed in the paper. The evaluation is based on simulated network parameters and realistic models.

Key Insights and Ideas

  • Practical Power Models Matter: Using realistic power consumption models for IoT devices significantly alters the latency-energy trade-off curve (Pareto boundary) compared to simplified or unrealistic models. This highlights the importance of accurate modeling for practical system design.
  • NBS for Fairness and Optimality: The Nash Bargaining Solution provides a principled way to achieve both optimal resource allocation and fairness among competing IoT devices, outperforming simple resource sharing schemes.
  • Pareto Boundary as a Design Tool: The derived Pareto boundary is not just an academic exercise; it's a valuable tool for network designers, allowing them to select operating points that balance latency and energy based on specific application requirements.
  • Joint Optimization is Crucial: Optimizing radio (transmit power) and computing (CPU frequency) resources jointly is essential for achieving efficient latency-energy trade-offs, as these resources are interdependent.

Key Results

  • The proposed NBS-based solution consistently outperforms a classical virtualization approach where CPU frequency is equally shared and devices transmit at maximum power.
  • The energy consumption achieved by the proposed solution is significantly lower than the non-optimized network, especially when the weight for energy consumption in the objective function is higher (lower $\eta$).
  • The achieved service latency is also substantially reduced compared to the non-optimized network, particularly when the CPU frequency is increased.
  • Increasing the bandwidth ($B$) affects the range of the energy consumption interval but has a negligible impact on the service latency range.
  • The practical power consumption model leads to a more constrained and potentially more realistic latency-energy trade-off compared to unrealistic models.

Limitations and Future Work

  • Single-Cell Network: The current model focuses on a single-cell fog-assisted network. Extending this to multi-cell scenarios with inter-cell interference would be a natural next step.
  • Fixed Task Offloading: The paper assumes that devices offload their entire tasks. Future work could explore partial offloading strategies.
  • Dynamic Environment: The current model assumes static device locations and channel conditions. Incorporating dynamic environments and mobility would enhance its applicability.
  • Computational Complexity: While the block coordinate descent algorithm is efficient, further research into distributed or more scalable optimization techniques for very large networks could be beneficial.

Summary updated

Abstract

This work aims to present a joint resource allocation method for a fog-assisted network wherein IoT wireless devices simultaneously offload their tasks to a serving fog node. The main contribution is to formulate joint minimization of service latency and energy consumption objectives subject to both radio and computing constraints. Moreover, unlike previous works that set a fixed value to the circuit power dissipated to operate a wireless device, practical models are considered. To derive the Pareto boundary between two conflicting objectives we consider, Tchebyshev theorem is used for each wireless device. The competition among devices is modeled using the cooperative Nash bargaining solution and its unique cooperative Nash equilibrium (NE) is computed based on block coordinate descent algorithm. Numerical results obtained using realistic models are presented to corroborate the effectiveness of the proposed algorithm.
BibTeX
@article{shams2023joint,
  author = {Farshad Shams and Vincenzo Lottici and Zhi Tian and Filippo Giannetti},
  title = {{Joint Latency-Energy Minimization for Fog-Assisted Wireless IoT Networks}},
  journal = {IEEE Open Journal of the Communications Society},
  year = {2023},
}
Figure