Emerging use cases, such as the orchestration of heterogeneous domains or the interconnection of remote data centers, require the design of new network control, management, and orchestration architectures that are adapted to the requirements associated with end-to-end services. This heterogeneity, stemming from actual deployments, is related to the involved data and control technologies or due to network segmentation. It is acknowledged that, in these scenarios, the deployment of a single software- defined networking (SDN) controller may not be practical or may have fundamental limitations, in terms of either scalability, complexity, or interoperability. To address this trend, and as a means to cope with the complexity growth, research in this area is considering the concept of SDN orchestration (the coordinated control of heterogeneous systems), abstraction (the selection of entityrelevant attributes), and the use of multiple controllers, commonly arranged in hierarchical or flat configurations. In this paper, we consider a hybrid approach combining elements from distributed control with elements from centralized control, relying on the concepts of abstraction and aggregation and hierarchical traffic engineering (TE). We propose a multidomain and multivendor network control architecture organized in layers, the abstract network layer and the control-specific layer, resulting in a mesh of generic SDN controllers that use generalized multiprotocol label switching (GMPLS) protocols as their east/west interfaces (open shortest path first TE and Resource Reservation Protocol with extensions for TE as inter- SDN controller communications protocols), forming an abstracted topology. To validate the approach, and to evaluate factors such as end-to-end network service deployment delay or control overhead, the proposed architecture is implemented and validated in a control plane testbed. While emphasis is given to the control of optical (flexi-grid) networks, the proposed approach can be generalized to other transport technologies.
Read full abstract