Minimum-cost embedding of virtual networks: An iterative decomposition approach
Yang Wang,Qian Hu,Long Nguyen,Maryam Jalalitabar
DOI: https://doi.org/10.1016/j.comnet.2023.109907
IF: 5.493
2023-07-07
Computer Networks
Abstract:The ossification of the current Internet can be mitigated by network virtualization. The key idea of network virtualization is to abstract a logical service as a virtual network, which is mapped to the physical infrastructure as networked virtual machines. This process, known as virtual network embedding (VNE), entails two components: Node Assignment (NA), where virtual nodes are mapped to physical nodes as virtual machines (i.e., via node virtualization); and Link Mapping (LM), where physical paths are allocated to connect the embedded virtual nodes (i.e., via link virtualization). The existing VNE solutions based on Integer Linear Programming models and heuristic algorithms, unfortunately, fall in short in addressing either one of the two following perspectives: ILP-based models address NA and LM jointly , leading to optimal results but they suffer from extensive computation time; heuristic algorithms simplify the problem computationally but it lacks guaranteed solution quality. A large body of VNE solutions, particularly, adopt strategies like divide-and-conquer: NA and LM are resolved as two independent sub-problems in sequence . This paper is motivated by an important mathematical quest to better leverage this decomposition strategy: how can one identify the inherent link between the NA and LM sub-problems to obtain a decomposition approach without trading off solution quality? The major contributions of this work lie on three facets that fulfill this quest. First, based on a path-based VNE model, our primal–dual analysis reveals the process towards formulations of the NA, and LM sub-problems that preserve the link (as in ILP solutions). Second, we develop an iterative decomposition approach that employs merits from both worlds: the complexity of VNE (structurally and computationally) is simplified by decomposition while optimality or guaranteed closeness to the optimality is attained. Third, our evaluations confirm that the proposed schemes can achieve trade-offs in time and optimality, and outperform benchmark schemes by a large margin.
computer science, information systems,telecommunications,engineering, electrical & electronic, hardware & architecture