Enhanced capacitated next controller placement in software-defined network with modified capacity constraint

Software-defined networking (SDN) is an emerging networking architecture paradigm that decouples the control and data planes. The problem of figuring out the number and positions of controllers and mapping of switches to them is known as the controller placement problem. To provide the resilience against the failure of a controller, each switch is mapped to a primary controller (first reference controller or FRC) and a backup controller (second reference controller or SRC). An existing work aims to minimize the worst-case latency (WCL) from switch to controller when a controller fails. But this work misses the constraint specifying the definition of a switch's SRC, which might cause an increase in the latency between some switches and their controllers in the event of a controller failure. In order to address this issue, a model is proposed in this paper by incorporating the missing constraint. But the addition of this constraint can potentially cause an increase in the minimum number of required controllers. In order to address this issue, a second model is proposed in this paper by modifying the capacity constraint based on the observation that the capacity of a controller need not be reserved for all the switches for which it acts as SRC. The two proposed models aim at minimizing the WCL from switch to controller when a controller fails. Three network topologies are used to test the proposed models and compare their performance with the existing model in terms of principal and subsidiary metrics. The results demonstrate that the proposed models perform on equal level with the existing model in terms of WCL from switch to SRC while outperforming it in terms of average latency (AL). For example, the first proposed model achieves an average AL reduction of 21.63%, 8.55%, and 25.13% compared with the existing model on three networks. Similarly, the second proposed model achieves an average AL reduction of 21.3%, 8.55%, and 24.19% in each network on three networks. Moreover, the second proposed model achieves a fair trade-off between the minimum number of controllers required and AL while outperforming both the existing and the first proposed models in terms of the average percentage of reserved controller capacity. This paper proposes two models that outperform the existing model in terms of average latency (AL). Moreover, the second proposed model achieves a fair trade-off between the minimum number of controllers required and AL when compared with the existing model, and it also outperforms both the existing and the first proposed models in terms of the average percentage of reserved controller capacity.image