Abstract
This paper introduces a unified "3C" framework for the Tactile Internet, enabling mobile edge devices to share Communication, Computation, and Caching resources through device-to-device (D2D) links. By transitioning from task-centric to resource-centric sharing, the proposed model optimizes energy consumption across heterogeneous devices. Simulation results indicate that this cooperative approach can reduce total energy usage by up to 83.8% while adhering to the stringent latency requirements of real-time edge applications.
Introduction
The "Tactile Internet" represents the next frontier of mobile connectivity, promising real-time remote control of edge devices such as robots and VR headsets. These applications demand ultra-low latency and extreme reliability, placing immense pressure on the limited resources of individual mobile devices. While previous research has explored sharing single resources (1C)—such as internet tethering or computation offloading—these models are typically "task-centric," creating silos where devices can only cooperate with others performing identical tasks.
This paper addresses these limitations by proposing a resource-centric model that breaks down these silos, allowing for the joint optimization of Communication, Computation, and Caching (3C) resources. This synergy enables a more efficient use of the collective edge pool, significantly reducing energy consumption without compromising critical deadlines.
The Unified 3C Framework
The core innovation of this research is the decomposition of every edge task into three fundamental subtasks: inputting (downloading or retrieving from cache), computation (CPU processing), and outputting (uploading or local caching). By treating these as a unified pool of resources, the framework allows a device with a high-performance processor but poor connectivity to assist a device with a fast internet connection but limited battery or CPU power.
As shown in the table below, this model is flexible enough to encompass various existing edge scenarios, from simple ad hoc content sharing to complex distributed data analysis.
Optimization Strategy and Heuristics
The authors formulate the resource allocation problem as an energy minimization challenge. Initially, this is an integer non-convex optimization problem due to the complex quadratic forms of the delay constraints. To resolve this, the authors employ a linear transformation technique to convert the non-convex constraints into linear ones, resulting in an Integer Linear Programming (ILP) problem.
Given that ILP is NP-complete and computationally expensive for large networks, a heuristic algorithm based on LP-relaxation is proposed. This algorithm iteratively solves a simplified version of the problem and prevents task-to-device allocations that violate delay bounds, ensuring a feasible solution is reached quickly.
Evaluation and Key Results
The framework was evaluated using numerical simulations on a network topology modeled as a random graph ($G(N, p)$). Device capacities and task demands were generated using truncated normal distributions based on real-world benchmarks for WiFi Direct and cellular energy consumption.
- Significant Energy Savings: The 3C framework achieved an 83.8% reduction in total energy consumption compared to traditional 1C/2C models when D2D energy costs were negligible. Even in scenarios where D2D transmission was twice as energy-intensive as cellular, savings remained substantial at 27.5%.
- Algorithmic Efficiency: For a network of 27 devices, the proposed heuristic was 78.6% faster than the optimal ILP solver, with an optimality gap of only 11.2%.
- The Power of Heterogeneity: A key finding is that the framework performs better as device diversity increases. High variance in device capacities allows the system to more effectively "cherry-pick" the most efficient resource for each specific subtask.
Conclusion
By shifting to a resource-centric 3C model, edge networks can unlock massive energy efficiencies while meeting the strict requirements of the Tactile Internet. Future work will focus on developing distributed algorithms to remove the need for a centralized scheduler and exploring proactive caching strategies to further minimize latency in dynamic mobile environments.
Summary updated