Isolation Scheme for Virtual Network Embedding Based on Reinforcement Learning for Smart City Vertical Industries

Modern ICT infrastructure is built on virtualization technologies, which connect a diverse set of dedicated networks to support a variety of smart city vertical industries (SCVI), such as energy, healthcare, manufacturing, entertainment, and intelligent transportation. The wide range of SCVI use cases require services to operate continuously and reliably. The violation of isolation by a specific SCVI, that is, a SCVI network must operate independently of other SCVI networks, complicates service assurance for infrastructure providers (InPs) significantly. As a result, a solution must be considered from the standpoint of isolation, which raises two issues: first, these SCVI networks have diverse resource requirements, and second, they necessitate additional functionality requirements such as isolation. Based on the above two problems faced by SCVI use cases, we propose a virtual network embedding (VNE) algorithm with resource and isolation constraints based on deep reinforcement learning (DRL). The proposed DRL_VNE algorithm can automatically adapt to changing dynamics and outperforms existing three state-of-the-art solutions by 12.9%, 19.0% and 4% in terms of the acceptance rate, the long-term average revenue, and long-term average revenue to cost ratio.