Reflections on Data Integration for SDN.

Software-defined networking (SDN) provides an unprecedented opportunity to exercise computing principles in networking practice. This paper investigates data integration, a under-explored discipline from the database community. We propose to manage the various SDN control applications that collectively drive a shared dataplane using a data integration system. First, we develop a baseline design and study its feasibility on two networking challenges not adequately addressed in classic data integration systems: the extensibility requirement to add new controls on demand, and the performance requirement to cope with fast dataplane updates. Central to our baseline design is a relational model, where the entire SDN is represented as relational data (values in tables) with distinct roles. The control applications act as data sources generating network state, and the dataplane becomes the integrated whole. Based on this model, we explore extensions to data integration systems called behavioral dependency, a formal notion that captures the dynamic interactions between the control applications. While our design and extension are not intended to provide a comprehensive solution, we believe our study is a step toward reaping the benefits of data integration for SDN.