Software-Defined Networking (SDN) potentially can improve the flexibility and management of Wireless Sensor Networks (WSNs). To investigate the impact of SDN on WSN, in this thesis, we consider three Software-Defined Wireless Sensor Network (SD-WSN) frameworks, namely SDN-WISE, SDWN-ONOS, and TinySDN. After comparing these frameworks, the performance of TinySDN is evaluated in three different scenarios: homogeneous, heterogeneous, and dynamic networks. The Collection Tree Protocol (CTP) is used for the evaluations, and is subsequently compared with Rime data collection protocol in Contiki. Our performance evaluation is based on four metrics: Packet Delivery Ratio (PDR), packet duplication, duty cycle, and delay.

Our results show that the PDR for WSN and SD-WSN is relatively similar, that is, around 0.98 to 1 for homogeneous and heterogeneous networks, whereas in a dynamic network, the PDR of WSN decreases from 0.98 to 0.9. Compared to the WSN, the SD-WSN reduces both the average delay of SDN sensor node and the time the SDN sensor node is active to send packets in homogeneous and heterogeneous networks. However, implementing a centralized controller in a dynamic network may cause the SD-WSN performance decrease, which is indicated by the increase ratio of average delay from 2.03 to 2.3, whereas the increase ratio of average delay for the WSN is only around 1.6 to 1.98. The packet duplication level increases by 33% in the dynamic network when the number of SDN sensor nodes increases from 10 to 15.

The performance of SD-WSN in heterogeneous, homogeneous, and dynamic networks is relatively worse than WSN in terms of packet duplication and Rx duty cycle. SD-WSN, in addition, is not optimally implemented for dynamic conditions as the activity changes from the SDN sensor nodes will significantly affect the performance of SD-WSN. To reduce the load on the centralized controller, the clustering controllers are deployed to distribute the load on multiple controllers. The result shows that packet duplication and average delay in a dynamic network can be reduced by 13% and 57%, respectively. Clustering controllers provide more stability in terms of Rx duty cycle compared to before using clustering controllers.