MBL-DSDN: a novel load balancing algorithm in distributed software-defined networks based on micro-clustering and B-LSTM methods

MBL-DSDN: a novel load balancing algorithm in distributed software-defined networks based on micro-clustering and B-LSTM methods

연구 분야: Networking

학회: The Journal of Supercomputing

While software-defined networks (SDNs) were introduced to enhance the performance of traditional networks, they brought forth significant challenges, such as lack of scalability, low reliability, and the presence of a centralized controller as a single point of failure. This led to the emergence of distributed software-defined networks (DSDNs). Despite the numerous advantages of DSDNs, there are challenges, with improper traffic distribution identified as the primary issue, significantly affecting its performance and efficiency, resulting in increased latency, reduced throughput, and migration costs. This article introduces a novel approach called MBL-DSDN (micro-clustering bidirectional LSTM-DSDN) to balance server traffic distribution. The improvement of quality of service (QoS) parameters and optimal load distribution in DSDNs is accomplished by utilizing two modules: micro-clustering (MC) and a prediction module based on the B-LSTM method. The Davies–Bouldin (DB) index from the MC module is utilized to enhance the accuracy of clustering results. Since the B-LSTM prediction module possesses suitable memory to store all relevant past and future features with high precision, it is suitable for selecting the best controller with the lowest migration cost and response time. Moreover, the proposed method reduces the synchronization cost between data and controller layers, leading to reduced response time and, ultimately, lower processing costs and increased load balancing. In comparison with similar methods, as a result of multiple simulations, migration count, migration cost, and response time parameters have improved by 31%, 43%, and 40.5%, respectively.