On the Transition of Legacy Networks to SDN - An Analysis on the Impact of Deployment Time, Number, and Location of Controllers
Marcos Fagundes Caetano, Geraldo P. Rocha Filho, Lisandro Zambenedetti Granville, Marcelo A. Marotta · Integrated Network Management · 2021
SDN has emerged as an alternative networking paradigm separating the control plane from the data plane. In real-world scenarios, the transition from a traditional network to an SDN one is usually done in incremental steps, due to the costs and bureaucracy of new hardware’s acquisition and deployment in companies. This transition involves the deployment of SDN-ready switches and hardware to run the software responsible for managing the control plane, i.e., the controller. Many studies on how to properly place and reduce the number of controllers, known as the Controller Placement Problem (CPP), have been conducted, but only a few of them exploited the CPP on hybrid (traditional/SDN) networks. In hybrid networks, we identified a gap in the minimum number of controllers required for a hybrid SDN to operate, considering a different number of transition steps and the heuristic to place these controllers within the network. Therefore, in this paper, we propose three heuristic approaches based on graph invariant for choosing the node to place a new controller reducing its number on a network that starts legacy, becomes hybrid, and ends fully SDN-enabled at the end of a finite transition step horizon. The proposed heuristics are compared against an optimized model and have been proven useful in scenarios where the network administrator does not have full control over the network evolution. Also, we found that the optimal multi-step network upgrade requires, on average, the same amount of controllers of a one-step migration, but it has to have full knowledge on each transition change of the hybrid network. Finally, we show that the performance of the heuristic approaches is related to the network topology and characteristics, such as traffic and physical switches distribution.