Scalable design of SDN controllers for optical networks using federation-based architectures
João Ricardo Borges dos Santos · 2016
The introduction of software-defined networking (SDN) principles in the optical communications ecosystem promises to change the way things are done nowadays. Although many of these principles still require proper maturation, the main premises behind SDN are quite attractive: 1) The separation of control and data plane will ultimately result in simpler hardware elements in the network with the network intelligence all concentrated in software controllers; 2) With this separation, the control plane is relieved of the hardware limitations intrinsic to the network equipment. This eases the creation of automated and virtualized functions at the controller where capacity can be added; 3) Open and standardized interfaces between applications, controllers, and network elements is going to promote innovation, interoperability and disaggregated systems. In this context, the network controller, whose main responsibility lies in the efficient configuration of the network equipment, is a crucial element to the SDN architecture. Typically, such controllers are centralized elements that detain the complete knowledge of the network. However, for large-sized networks, such concentration can lead to lengthier operations, heavier resiliency schemes, and availability of excessive storage and processing capacity. To obviate these issues and promote scalability, this paper presents a new controller architecture based on a federation of multiple sub-network controllers, each managing only a section of the network, conveniently coordinated by a hierarchically-superior controller. A clustering algorithm, based on single-linkage, is developed to create sub-networks under different grouping policies. In order to assess the impact of employing the federated approach in an optical network, we use a reference topology to determine the amount of routed demands and regeneration cards required. The results obtained show that the performance of the federated architecture can only resemble that of the single-controller scheme when sufficient path diversity is offered. Such diversity is mostly dependent on the number of routes computed between sub-networks and the sub-network cardinality. In addition, we also show that the traffic load and the grouping policy can negatively affect the routing performance.