Novel resource allocation mechanism for SDN-based data center networks

Abstract

With the development of virtualization technologies, Service Providers (SP) can offer services to customers by renting various kinds of hardware resources from Data Centers (DC) instead of purchasing hardware equipment frequently, which not only increases the flexibility, but also decreases the cost. For the multi-tenant DC environment, each SP can be regarded as a DC tenant. Then, the request of each tenant could be abstracted as a Virtual Data Center (VDC) which is implemented by mapping the required virtual nodes and links onto physical servers and optical paths. Such process is also referred to as the VDC mapping, during which the corresponding resources will be allocated. It is aware that the VDC mapping problem is essentially different from the virtual network mapping problem, and the former one is much more complex than the latter one. Therefore, the solution of VDC mapping still faces many challenges. In this work, we focus on addressing the VDC mapping problem with a flexible and efficient resource allocation. Firstly, the concepts of Software Defined Networking (SDN) and Network Function Virtualization (NFV) are both introduced to construct a logically centralized and physically distributed resource allocation framework which decouples the server resource control and the network resource control. In this way, the whole resource allocation process can be managed in a more flexible and targeted way towards achieving high resource utilization. Secondly, a location-aware VDC partition process is introduced before VDC mapping based on the label propagation algorithm, which clusters the virtual nodes of a VDC request according to the communication frequency between any two virtual nodes. Generally, the higher the communication frequency, the more cost will be spent on network resource consumption, such that the virtual nodes clustered in the same group will be mapped onto the same DC to avoid excessive backbone network consumption. Lastly, a cost and security aware VDC mapping mechanism is proposed to address the corresponding resource allocation problem, which aims at minimizing the overall cost with both the basic resource constraints, mapping constraints, the wavelength continuity constraint and the security constraint taken into consideration. The experiment indicates that the proposed mechanism can decrease the overall VDC mapping cost by about 10%–25%, and increases the VDC acceptance rate and the total revenue by up to 34% and 71% respectively compared to the state-of-the-art benchmarks.