Research On Models And Algorithms Of Energy Efficient Virtual Network Mapping

The resources of the current Internet and the cloud data centers are designed for the peak loads. Excess supply of the substrate resources ensures the normal operation of the services. However, the low utilization of the substrate resources results in the huge waste of the electrical energy. Network virtualization is an important technology for the future Internet, the cloud data centers and the software-defined networks, and is an enabler for intelligent energy-aware network deployment. By means of the resource consolidation strategy, the network virtualization uses electric energy rationally and effectively, which enables the intelligent energy to deploy network efficiently. To construct the green networks and the green cloud data centers by the technology of the network virtualization has been one of the important research hot point.Energy-aware virtual network embedding (VNE) has been one of the major challenges and a critical research in the environment of the network virtualization. More attentions of the academics have been paid to the research on the energy efficient VNE (EEVNE). Lots of models and algorithms of VNE have been proposed, most of them focus on the cost minimization of VNE. Due to without considering the substrate network (SN) energy consumption, the huge unnecessary energy consumption has been produced by those mod-els and algorithms of VNE. Recently, several EEVNE models and algorithms have been proposed, but the following aspects need to improve:1) The existing EEVNE models and algorithms mainly use the resource consolidation strategy, but the energy consumption of SN is still high; 2) The existing EEVNE models and algorithms mainly focus on reduc-ing the energy consumption under the non-periodic dynamic loads, the research scope of EEVNE is limited. Based on the aforementioned reasons, the paper focuses on the EEVNE to resolve the above two problems, and makes the following contributions:1 To reduce the energy consumption of SN and the time complexity of the multi-commodity-flow-based virtual link mapping algorithms, we propose the virtual link embedding algorithms for the splitting path based on the minimum cost flow model. In the virtual link embedding, we find the dynamic inversion phenomenon and explain the reason of the phenomenon. The experimental results show that the dynamic inversion phenomenon is objective in the VNE, and the minimal cost flow based algorithm with low time complexity can reduce the energy consumption.2 Due to the close relationships between the energy consumption and the utilization of the substrate node, we propose an EEVNE transport model, whose objective is to minimize the energy consumption of the virtual node mapping. We propose an EEVNE algorithm based on the smallest element, which achieves the minimization of the energy cost under the constraint of the virtual link embedding. We also propose other transport-model-based EEVNE algorithm, where the different virtual nodes of the same VN can be mapped in the same substrate node. We also analyse their effects on the energy consumption. The experimental results show that the algorithms can reduce energy consumption efficiently.3 Using the strategy of the resource consolidation and minimizing the energy consumption cost of the virtual node embedding, we propose the multi-objective decision model for EEVNE. For the sake of solving the model, we utilize the scalar method to transfer the multi-objective decision model to the Interger Linear Programming model. A GLPK-based program is designed to resolve the integer linear programming model. We implement the simulations of the different EEVNE models in the topology of the future internet and the cloud data centers. The experimental results show that the model can reduce the energy consumption effectively. An enhanced EEVNE algorithm based on the smallest element method is proposed, which combines the resource consolidation strategy and the minimization of the energy consumption cost of the virtual node embedding. The experimental results show that the heuristic algorithm can reduce the energy consumption efficiently.4 Due to the VN dynamical effects on the energy consumption of SN, we propose two ac-tively hibernating substrate resources based methods, which can reduce the frequent switchs between the active state and the sleepy state in the substrate resources. In the dictionary-database-based method, we build a dictionary database for VNE, an algorithm is presented which trains the node and link utilization from the database, besides, a method is designed to look for the set of the hibernating resources. In the feedback-control-based method, we utilize the feedback relationship among VN requests, and propose a novel method, which extends the area of the substrate hi-bernating resources. We do the experiments in the synthetic network topology and USNET network topology, and in the different link energy consumptions where the en-ergy consumption of USNET is related to the link length but the energy consumption of the synthetic network is not related to the link length, respectively. The experi-mental results show that the proposed method can reduce the energy consumption obviously.5 Since the dynamical change of the substrate loads is periodical, we study EEVNE under this environment. We propose a multi-feedback control method for EEVNE under the environment of the above change of the substrate loads, and propose an algorithm to simulate periodically dynamic loads. The experimental results show that the proposed method can not only reduce the energy consumption evidently, but also increase the revenue and VN acceptance ratio.