Research On Efficient Embedding And Traffic Management For Network Virtualization

Network Virtualization has been propounded as an effective way to promote inno-vations in the current Internet architecture. There are two main challenges in networkvirtualization, i.e., efficient embedding for virtual networks and traffic management forphysical networks. On one hand, each virtual network may span over multiple physicalnetworks, resulting difficulties in designing global optimal embedding decisions. It isalso hard to abstract resource requirements for virtual networks in data center networks.Therefore, high-efficient virtual network embedding mechanisms are extremely requiredto satisfy resource requirements for virtual networks and to improve resource utilizationfor physical networks. On the other hand, superlinearly increasing of network trafficleads to a higher risk of physical network congestion, as well as an ever-growing energyconsumption of network equipments. Traffic management mechanisms are thus in needto optimize network performance and lower the overall energy consumption.To address those challenges, this paper aims to improve embedding efficiency forvirtual networks and traffic management performance for physical networks. With thesegoals, we focus on4specific research issues, including inter-domain virtual network em-bedding, virtual resource allocation in data center networks, generalized routing frame-work for optimal traffic engineering, and energy-efficient routing mechanisms.The main contributions of this paper are summarized as follows:1. Designing an inter-domain virtual network embedding mechanism. Consider-ing the information asymmetry in the embedding process, we first propose a limited in-formation sharing scheme involving multiple physical domains, and then formulate theembedding problem as an integer optimization problem. Finally, we devise an efficientpolynomial-time algorithm for online embedding requests. Through extensive experi-ments, we show that the performance of the proposed mechanism, in terms of successratio and physical resource utilization, can achieve80%–90%of the ideal results.2. Designing a virtual resource allocation mechanism with bandwidth guaranteesfor data center networks. To address the difficulty in describing resource requirementsfor virtual networks, we advocate decoupling bandwidth guarantees for different types ofcommunications, and then propose a novel Dual-Hose model as well as corresponding VM and bandwidth allocation algorithms. Experiment results show that, comparing withthe state-of-the-art mechanism, the newly-proposed one can improve the request accep-tance ratio by5.3%.3. Proposing a generalized routing framework for the optimal traffic engineering(TE). To satisfy diverse TE objectives and implementation requirements, we first intro-duce the network utility maximization model to obtain a generalized routing framework,then propose a general load-balancing utility function, and finally design a new link-state routing protocol named SPEF that can realize the optimal routing. In addition, weanalyze the MPLS-based implementation. Experiment results show that1) the networkutility with SPEF is10%–40%larger than that with the commonly-used OSPF protocol,and2) only6.67%of the optimal routes in MPLS-based implementation varies over time,which confirms its stability.4. Designing ROD (Routing On Demand), a reliable energy-efficient routing mech-anism. To address the interaction of multiple TE objectives including performance,energy-efficiency and routing robustness, ROD resort to a hierarchical combination opti-mization. We first formulate an optimization problem by jointly considering performanceand energy-efficiency, which results in the optimal routing under specific constrains.Then, we optimize the robustness of the derived routing over all single link failures.Through experiment results, we show that ROD can get the desired routing by flexiblyadjusting trade-offs among those three TE objectives.