Scholarlink

<p>High-performance computing is in great demand in resource-constrained environments (e.g., Mobile Edge Computing (MEC) and Internet of Things (IoT) systems), making it necessary to have effective resource allocation strategies. In this paper, we propose a kind of elite approach that can combine Multi-Agent Reinforcement Learning (MRL) and Deep Reinforcement Learning (DRL) to optimize task offloading as well as resource management in MEC environments. Existing approaches suffer from non-negligible computational latency, inefficient energy consumption, and limited resource utilization under varying network and computation loads. Our methodology for confronting these challenges combines an adaptive resource allocation mechanism, which dynamically balances network bandwidth and CPU resource allocations according to task requirements in real-time. The proposed method is evaluated across several key metrics, including convergence performance, computation latency, energy consumption, and resource utilization rates. Experimental results show significant improvements over state-of-the-art methods. The proposed approach achieves a convergence performance of 1.00 at 250 iterations, which is superior to DRL (0.97), MRL (0.90). The proposed method achieves 0.068&#xa0;s, outperforms the DRL (0.083 ms) and MRL (0.165 ms) in terms of computation latency, with an input size is 0.2 Mbits. The energy consumption is also reduced, where 0.038 Joules for 0.2 Mbits have been consumed in the proposed method; this value is much smaller compared to DRL (0.044 Joules) and CERAI (0.132 Joules). These results show that the proposed method is superior in a balanced performance, energy efficiency, and resource utilization perspective for MEC dense areas with limited resources.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid deep reinforcement learning and genetic algorithm-based resource allocation framework for edge computing

  • 期刊论文

  • 发表日期：

    2025年10月1日

查看全文
  • V. Arun,
  • M. Azhagiri

摘要

High-performance computing is in great demand in resource-constrained environments (e.g., Mobile Edge Computing (MEC) and Internet of Things (IoT) systems), making it necessary to have effective resource allocation strategies. In this paper, we propose a kind of elite approach that can combine Multi-Agent Reinforcement Learning (MRL) and Deep Reinforcement Learning (DRL) to optimize task offloading as well as resource management in MEC environments. Existing approaches suffer from non-negligible computational latency, inefficient energy consumption, and limited resource utilization under varying network and computation loads. Our methodology for confronting these challenges combines an adaptive resource allocation mechanism, which dynamically balances network bandwidth and CPU resource allocations according to task requirements in real-time. The proposed method is evaluated across several key metrics, including convergence performance, computation latency, energy consumption, and resource utilization rates. Experimental results show significant improvements over state-of-the-art methods. The proposed approach achieves a convergence performance of 1.00 at 250 iterations, which is superior to DRL (0.97), MRL (0.90). The proposed method achieves 0.068 s, outperforms the DRL (0.083 ms) and MRL (0.165 ms) in terms of computation latency, with an input size is 0.2 Mbits. The energy consumption is also reduced, where 0.038 Joules for 0.2 Mbits have been consumed in the proposed method; this value is much smaller compared to DRL (0.044 Joules) and CERAI (0.132 Joules). These results show that the proposed method is superior in a balanced performance, energy efficiency, and resource utilization perspective for MEC dense areas with limited resources.