Loading...
Search
Search in this resource
sort by
Send to Friend
 Digital Object List
 Bookmark
 Review List
  1. مدیریت توامان بارسپاری محاسبات و ذخیره سازی خدمات در محیط محاسبات لبه

Joint Computation Offloading and Service Caching in Edge Computing Environments

Yavari Darani, Amir Arsalan | 2025

9 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58892 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Hessabi, Shaahin; Movaghar, Ali
  7. Abstract:
  8. With the expansion of the Internet of Things (IoT) and the growing need for real-time data processing, cloud computing has been adopted as a solution for convenient access to large-scale computational resources. However, transmitting massive volumes of distributed data to the cloud can lead to increased network congestion, latency, and reduced efficiency. In response to this challenge, edge computing has emerged with the aim of performing computation and storage closer to data-generation sources, attracting the attention of both researchers and industry practitioners. The proximity of edge nodes to IoT devices reduces latency and improves the utilization of data; nevertheless, the limited processing and storage resources of these nodes, along with the need to provide services tailored to different applications, makes their efficient management a serious challenge. Consequently, improper application of computation offloading and storage policies can prevent the full exploitation of the available capacity. The joint optimization of the aforementioned aspects in an edge-cloud environment belongs to the class of mixed-integer nonlinear programming (MINLP) problems. Therefore, one effective approach to solving this problem is the use of machine learning. On the other hand, the scarcity of training data and the difficulty of designing comprehensive and dynamic models limit the applicability of traditional methods. In recent years, deep reinforcement learning (DRL) has attracted attention as a novel approach. Considering the distributed nature of edge nodes and privacy-related challenges, centralized solutions may give rise to several issues. Therefore, this research proposes a distributed hierarchical collaborative deep reinforcement learning approach that achieves better large-scale performance compared with conventional centralized deep reinforcement learning methods. Simulation results show that, by jointly optimizing energy consumption and delay, the proposed approach stabilizes model performance by reducing fluctuations in the reward curve during the final training rounds, reduces network congestion by approximately 56%, and simultaneously improves users' privacy in the learning process more effectively than previous works that solved the problem using centralized deep reinforcement learning
  9. Keywords:
  10. Edge Computing ; Task Offloading ; Hierarchical Federated Learning ; Deep Reinforcement Learning ; Optimization ; Service Caching

 Digital Object List

 Bookmark

...see more