Feature Selection Strategies for the Controller Placement Problem in SDNs: A Review

Recent advances in computing makes it possible for a myriad of devices, people and things to connect together irrespective of space and time constraints. However, for this to happen next-generation networks need to be fast, scalable, reliable and cost effective. Interestingly, Software defined networking (SDN) provides the enabling paradigm for this to happen. However, the placement of controllers in this emerging paradigm is crucial to its performance and scalability hence its relevance. To this end, we first present a concise review of four controller placement models relevant to this study, highlighting their objectives, strengths and gaps. We further illustrate how the feature selection strategies used by these models affect their accuracy and convergence times, thus making them unsuitable for real life scenarios. Although three out of four of the reviewed models consider traffic conditions in placing controllers in SDNs, they nonetheless incur high convergence times. There is, however the possibility of achieving high accuracy with low convergence by fine tuning the feature selection techniques implemented by these models. The deliverable in view is a road map towards the development of an optimized controller placement model where optimality means maximizing the model's accuracy whilst minimizing its convergence time. In this way we will be providing the necessary mechanism for an optimized controller placement model that responds to changing traffic conditions.