چكيده به لاتين
With the increasing growth of computer networks and the complexity of communication structures, the need for optimal traffic management, data security assurance, and reduction of operational and maintenance costs has become fundamental challenges in this field. In this context, Software Defined Networking (SDN) architecture, as a novel paradigm in modern network management, has been able to provide flexibility, scalability, and centralized management capabilities through the separation of control and data planes. In SDN, controllers act as primary elements of network management and decision-making, responsible for tasks such as traffic monitoring, resource allocation, and quality of service provision. One of the fundamental challenges in SDN architecture is the optimal placement of controllers. This challenge arises because the distribution and deployment of controllers directly affects overall network performance, latency, load balancing, scalability, and security. In centralized architectures, a single controller acts as a single point of failure, causing network disruption if it fails. On the other hand, using multiple distributed controllers can largely resolve this issue, but raises new concerns such as coordination between controllers, increased latency due to traffic redirection, and communication overhead between controllers. Additionally, network dynamics, continuous changes in traffic, and the need for real-time responsiveness to network requests further complicate the controller placement problem. Previous research in this area has primarily addressed specific aspects of the placement problem and paid less attention to multi-objective optimization. In this study, a multi-objective approach based on machine learning and specifically reinforcement learning is proposed for optimal controller placement in SDN networks. This method considers criteria such as latency reduction, load balancing, stability assurance, and operational efficiency improvement, and utilizes reinforcement learning algorithms to search for optimality in the problem space and adapt to high network dynamics. The significance of this research lies in the fact that considering the NP-Hard nature of the controller placement problem, traditional approaches for solving this problem are time-consuming and inefficient. Reinforcement learning algorithms can identify and deploy optimal controller locations under various network conditions by leveraging feedback-based learning processes. In this research, various performance metrics have been optimized simultaneously, and evaluation results show that the proposed method has achieved significant improvements compared to previous methods in terms of parameters such as latency reduction, load balancing, and network stability. The main innovation of this research is its special focus on stability as one of the key objectives in the controller placement process, as well as achieving an average 7% improvement in network efficiency compared to conventional methods. This achievement demonstrates that the proposed method can be used as an efficient framework for simultaneous optimization of multiple key objectives in SDN networks. In general, this investigation not only presents an innovative framework for optimal controller placement but also offers high adaptability and flexibility in dealing with network dynamics through reinforcement learning.
