AN INTELLIGENT LOAD BALANCING STRATEGY TO IMPROVE PERFORMANCE AND QoS IN SD-DCN (SOFTWARE DEFINED-DATA CENTER NETWORK)

SDN (Software-defined Networking) is a new state-of-the-art architectural approach to network management. It enables more flexible management of large-scale, complex networks such as data center networks. To improve data transmission performance in SD-DCN, this paper proposes a strategy for intelligent load balancing of links in the network through machine learning. Thus, we use a comprehensive SDN method to assess the state of the network by examining switch load and link bandwidth utilization. In the DCN network, our algorithm uses two classification algorithms (Random Forest XGBoost) to classify elephant and mouse flows, enabling adaptive learning to the load balancing module consisting of a Deep-Q Learning (DQN) agent combined with one of the convolutional neural networks (CNN). By improving network efficiency and reducing packet loss, our Flow Classification and Optimized Path Prediction Algorithm (FCOPPA) is able to create optimal routing paths based on current network state and traffic data. The effectiveness of our algorithm is confirmed by simulations carried out in a Mininet environment with the RYU controller, using a fat-tree data center topology. The results show that it performs better in achieving higher throughput, lower latency and more efficient load balancing than conventional algorithms such as equal-cost multipath (ECMP) and Hedera. Received: July 23, 2024Revised: August 23, 2024Accepted: August 30, 2024