Online virtual network function placement in 5G networks

The placement of network functions in 5G networks, known as the Virtual Network Function-Forwarding Graph Embedding (VNF-FGE) problem, presents challenges in resource allocation, energy efficiency, and real-time service delivery. This paper introduces two reinforcement learning methods, OGA and Hybr...

Full description

Saved in:
Bibliographic Details
Published in:Computing Vol. 107; no. 5; p. 120
Main Authors: Esfandyari, Alborz, Zali, Zeinab, Hashemi, Massoud Reza
Format: Journal Article
Language:English
Published: Vienna Springer Vienna 01.05.2025
Springer Nature B.V
Subjects:
5G mobile communication
Artificial Intelligence
Computer Appl. in Administrative Data Processing
Computer Communication Networks
Computer Science
Energy consumption
Genetic algorithms
Information Systems Applications (incl.Internet)
Linear programming
Optimization
Performance evaluation
Placement
Real time
Regular Paper
Resource allocation
Software Engineering
Virtual networks
Wireless networks
Multi-access edge computing
68
Genetic algorithm
49
Optimal placement
Reinforcement learning
5G wireless network technology
Network function virtualization
94
ISSN:0010-485X, 1436-5057
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The placement of network functions in 5G networks, known as the Virtual Network Function-Forwarding Graph Embedding (VNF-FGE) problem, presents challenges in resource allocation, energy efficiency, and real-time service delivery. This paper introduces two reinforcement learning methods, OGA and Hybrid-OGA, modeling the VNF-FGE problem as a binary linear programming problem. OGA uses a seq2seq model with actor-critic reinforcement learning for optimal service chain placement, while Hybrid-OGA incorporates a genetic algorithm for refining blocked services placement. These methods address critical issues in 5G network function placement, optimizing VNF placement to minimize energy consumption while maintaining service performance and dependability. We evaluate the methods against First-Fit and CPLEX optimization tools, showing competitive performance in response time, accepted ratio, blocked services, and objective function minimization. Our methods reduce the objective function by 15–40% and improve the accepted ratio by 5–20% compared to CPLEX. With increased request rates, our methods show a 7% decrease in the accepted ratio, while others show a 14–20% decrease. Additionally, our methods increase the objective function by 20–33%, compared to a 48% increase in methods with higher service block rates.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0010-485X
1436-5057
DOI:10.1007/s00607-025-01474-3