FlagVNE: A Flexible and Generalizable Reinforcement Learning Framework for Network Resource Allocation

• DOI: 10.48550/arxiv.2404.12633
• Publication Date: 2024-04-19
• Authors: Tianfu Wang; Qilin Fan; Chao Wang; Long Yang; L. K. Ding; Nicholas Jing Yuan; Hui Xiong
• Keywords: Computer Science - Networking and Internet Architecture; Networking and Internet Architecture (cs.NI); FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence

Virtual network embedding (VNE) is an essential resource allocation task in virtualization, aiming to map virtual requests (VNRs) onto physical infrastructure. Reinforcement learning (RL) has recently emerged as a promising solution this problem. However, existing RL-based VNE methods are limited by the unidirectional action design and one-size-fits-all training strategy, resulting restricted searchability generalizability. In paper, we propose FLexible And Generalizable RL framework for VNE, named FlagVNE. Specifically, bidirectional action-based Markov decision process model that enables joint selection of nodes, thus improving exploration flexibility space. To tackle expansive dynamic space, hierarchical decoder generate adaptive probability distributions ensure high efficiency. Furthermore, overcome generalization issue varying VNR sizes, meta-RL-based method with curriculum scheduling facilitating specialized policy each size. Finally, extensive experimental results show effectiveness FlagVNE across multiple key metrics. Our code available at GitHub (https://github.com/GeminiLight/flag-vne).