Scholarlink

<p>Internet of Things (IoT) produces voluminous data, which necessitates efficient data management by utilizing computing resources deployed close to the network edge. This data management requires efficient techniques to offload IoT tasks/data, as task offloading is an NP-hard problem. The task offloading policies should address critical issues, such as resource limitations, load distribution, time involved, reliability, etc., in transferring tasks from the edge layer to the fog-cloud layer. Though a good amount of research has been done on task offloading, only a few considered the node’s resource occupancy prediction, which may guide the selection of appropriate computing nodes for task offloading. This work proposes a two-step method, called ITO-PDR, for task offloading on limited edge-fog-cloud resources. The first step predicts the future resource occupancy enabling proactive task scheduling and minimizing potential delays using bidirectional Long-short-term memory (BiLSTM). In the second step, deep reinforcement learning is applied to design a task-offloading policy based on the status of the network and the predicted outcomes. Performance evaluation shows that the proposed method reduces average task delay by approximately 30%, compared to baseline algorithms, and achieves successful task offloading rate by 95%. These results demonstrate the efficacy of the ITO-PDR in enhancing the Quality of Service (QoS) in IoT environments.</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

Intelligent Task Offloading: Reducing Delay and Offloading Failure Using Predictive Resource Allocation and Deep Reinforcement Learning (ITO-PDR)

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Pratibha Yadav,
  • Deo Prakash Vidyarthi

摘要

Internet of Things (IoT) produces voluminous data, which necessitates efficient data management by utilizing computing resources deployed close to the network edge. This data management requires efficient techniques to offload IoT tasks/data, as task offloading is an NP-hard problem. The task offloading policies should address critical issues, such as resource limitations, load distribution, time involved, reliability, etc., in transferring tasks from the edge layer to the fog-cloud layer. Though a good amount of research has been done on task offloading, only a few considered the node’s resource occupancy prediction, which may guide the selection of appropriate computing nodes for task offloading. This work proposes a two-step method, called ITO-PDR, for task offloading on limited edge-fog-cloud resources. The first step predicts the future resource occupancy enabling proactive task scheduling and minimizing potential delays using bidirectional Long-short-term memory (BiLSTM). In the second step, deep reinforcement learning is applied to design a task-offloading policy based on the status of the network and the predicted outcomes. Performance evaluation shows that the proposed method reduces average task delay by approximately 30%, compared to baseline algorithms, and achieves successful task offloading rate by 95%. These results demonstrate the efficacy of the ITO-PDR in enhancing the Quality of Service (QoS) in IoT environments.