Age-of-Information-Aware Distributed Task Offloading and Resource Allocation in Mobile Edge Computing Networks

Minwoo Kim,Jonggyu Jang,Youngchol Choi,Hyun Jong Yang

2024-04-04

Abstract:The growth in artificial intelligence (AI) technology has attracted substantial interests in age-of-information (AoI)-aware task offloading of mobile edge computing (MEC)-namely, minimizing service latency. Additionally, the use of MEC systems poses an additional problem arising from limited battery resources of MDs. This paper tackles the pressing challenge of AoI-aware distributed task offloading optimization, where user association (UA), resource allocation (RA), full-task offloading, and battery of mobile devices (MDs) are jointly considered. In existing studies, joint optimization of overall task offloading and UA is seldom considered due to the complexity of combinatorial optimization problems, and in cases where it is considered, linear objective functions such as power consumption are adopted. Revolutionizing the realm of MEC, our objective includes all major components contributing to users' quality of experience, including AoI and energy consumption. To achieve this, we first formulate an NP-hard combinatorial problem, where the objective function comprises three elements: communication latency, computation latency, and battery usage. We derive a closed-form RA solution of the problem; next, we provide a distributed pricing-based UA solution. We simulate the proposed algorithm for various vision and language AI tasks. Our numerical results show that the proposed method Pareto-dominates baseline methods. More specifically, the results demonstrate that the proposed method can outperform baseline methods by 1.62 times smaller AoI with 41.2% less energy consumption.

Signal Processing

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the Age-of-Information (AoI) aware distributed task offloading and resource allocation problem in Mobile Edge Computing (MEC) networks. Specifically, the paper focuses on the following key challenges:

1. **Complete Offloading of AI Tasks**: Most AI tasks are not suitable for partial offloading, thus requiring hard decisions rather than soft decisions in complex scenarios, which increases the difficulty of task offloading optimization.

2. **Joint Optimization of User Association (UA) and Resource Allocation (RA)**: In MEC networks, especially in dense or heterogeneous networks, multiple base stations are available for mobile devices (MDs) to connect. Each MD is associated with an edge server (ES) and receives allocations of computing and communication resources. However, due to the diversity in computing capabilities of local and edge devices and the presence of computation and communication delays, joint optimization of UA and RA becomes very complex.

3. **Trade-off Between AoI and Energy Consumption**: In MEC systems, MDs can choose to offload tasks to edge servers or process them locally. When the edge server is heavily loaded, local computation, despite consuming more energy, may have an advantage in terms of delay. Therefore, it is necessary to prioritize the remaining battery life of MDs to maintain network continuity and find effective solutions to balance these factors.

### Main Contributions of the Paper

1. **Proposed an Innovative AI Task Offloading Strategy**: This strategy not only prioritizes the battery levels of MDs but also achieves fully distributed load balancing optimization.

2. **Solved the Mixed Integer Combinatorial Optimization Problem**: By introducing a relaxation technique based on auxiliary variables, it allows balancing the load through continuous variables instead of combinatorial variables, thereby achieving distributed optimization and reducing execution complexity.

3. **Provided Detailed Analysis of Convergence and Optimality of the Distributed Optimization Problem**: These analyses provide theoretical support for the effectiveness of the method.

4. **Validated the Effectiveness of the Scheme Through Experimental Results**: Experimental results show that the proposed scheme outperforms baseline models in terms of delay and energy consumption. Specifically, the scheme can reduce delay by up to 5.29 times and energy consumption by 64.0% on certain objectives.

### Conclusion

This paper addresses the AoI-aware distributed task offloading and resource allocation problem in MEC systems by proposing a new optimization method. This method not only considers the battery life of MDs but also achieves efficient utilization of computing and communication resources, having significant practical application value.