Double agents-DQL based D2D computing-offloading for SHVC

To make the network provide a better Quality-of-Service (QoS) guarantee for compute-intensive applications, many researchers study Device-to-Device (D2D) based Multi-access Edge Computing (MEC) for reducing computing task completion time and energy consumption. However, the majority of works applies...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Peer-to-peer networking and applications 2022-01, Vol.15 (1), p.56-76
Hauptverfasser: Liu, Jianlong, Wen, Jiaye, Lin, Lixia, Zhou, Wen’an
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To make the network provide a better Quality-of-Service (QoS) guarantee for compute-intensive applications, many researchers study Device-to-Device (D2D) based Multi-access Edge Computing (MEC) for reducing computing task completion time and energy consumption. However, the majority of works applies only to the scenario, in which, a specific video coding algorithm with a fixed number of enhancement layer is employed, i.e., the computing task with certain computing amount. Then, the computing task can only be continuously executed by edge computing-offloading. In fact, the Scalable High-efficiency Video Coding (SHVC) algorithm has an uncertain computing amount. Also, the computing task has multiple execution steps by D2D computing-offloading. This is because the executor may interrupt the execution of the ongoing SHVC computing task and change its execution step. Therefore, it is a challenging problem that, how to minimize the completion time and energy consumption of this scenario. To solve this problem, firstly, we model the SHVC computing tasks by ten atomic operations. Secondly, we propose an action space dynamic generation method to solve the reward sparse problem, and we propose an action space coding method to reduce the storage space and searching cost of experience pool matrix, and we employ a double-agent deep Q-learning algorithm to further improve the efficiency of solving the problem. Finally, the simulation results show that the long-term QoS of users is improved.
ISSN:1936-6442
1936-6450
DOI:10.1007/s12083-021-01203-5