Optimizing the load balancing problem in software-defined networks

Recently, there has been considerable interest in traffic engineering in Software-Defined Networks. In particular, the issue of server-based and connection-based load balancing in the data plane is of interest. Server-based load balancing deals with load balancing between servers with the same specifications, while connection-based load balancing deals with determining the path with the least load among the paths between the source and the destination. Within the scope of the thesis, different approaches have been proposed for these two subjects, and their success has been proven. Routing flows with different characteristics with the same routing strategy in connection-based load balancing is unrealistic. For this reason, in the thesis study, elephant flows were determined by classifying the flows. The proposed framework for elephant flow detection has been tested with seven different classifiers, and simulation results have proved the success of the methods. When the results were examined, it was determined that Decision Trees, Support Vector Machines and Deep Learning methods produced more successful results for elephant flow detection. Within the scope of the thesis, the elephant flows are routed by the Discrete–Particle Swarm Optimization method, which dynamically balances load, which is recommended for the connection-based load balancing approach. Also, Mouse flows are routed by traditional Round-robin or Random methods. The most successful result was obtained from the method in which the mouse flows were routed according to Round Robin by performing elephant flow detection according to the Deep Learning method.