A balanced partitioning mechanism for multicontroller placement in software-defined wide area networks

• Main Author: Saab, Maher Waleed Asaad
• Format: Thesis
• Language: English; English
• Published: 2022
• Subjects: T58.5-58.64 Information technology
• Online Access: https://etd.uum.edu.my/10232/1/s901196_01.pdf; https://etd.uum.edu.my/10232/2/s901196_02.pdf; https://etd.uum.edu.my/10232/

Through softwarization, Software-Defined Networking (SDN) may govern the network. Deploying a single controller to manage enormous network traffic is inefficient; hence, having multiple controllers is a necessity of current SDN in wide area networks (WANs). However, the controller placement problem (CPP) is a thriving research subject for efficiently placing many controllers to improve network performance. It has two parts: how the controllers should be distributed and how many networking devices each controller should be connected to. Consequently, the objective of this study is to propose a Balanced Partitioning Mechanism (BPM) based on the notion of a network partition. Moreover, the BPM is designed based on a modified K-means algorithm. BPM comprises of two approaches: the initialization method and the partitioning strategy. The farthest-point initialization method is introduced to reduce end-to-end delay between the controllers and switches. The balanced partitioning strategy is used to balance controller loads and partition the network into balanced partitions. The research adopted the Design Science Research Methodology (DSRM) to accomplish its objectives. The network simulator OMNeT++ was configured to simulate the performance of BPM over the OS3E topology, with two scenarios including five and six domains. The K-means and CNPA algorithms, in particular, were used to evaluate the performance of BPM. In terms of balanced partitioning, the findings reveal that BPM outperforms the K-means and CNPA algorithms by maintaining a good load balance among controllers. Furthermore, the results show that BPM improves throughput and reduces end-to-end delay between the controllers and switches. In addition, BPM improves the number of packets received by the destination to the number of packets sent by 23% and 29% compared to the K-means for five and six domain scenarios, respectively. Given the diversity of future Internet and IoT, the findings have significant implications for improving the performance of WAN networks.