A Framework for Performance Analysis of Various Load Balancing Techniques in a Software-Defined Networking Environment

摘要

Load balancer is an essential part of a computer network. Its primary purpose is to distribute incoming traffic across multiple target servers. There are numerous load balancing techniques and each of them excels on specific network topology and server capability. However, due to vendor dependency, implementing a quintessential load balancer requires additional hardware cost and knowledge in vendor-specific configurations. Using software-defined networking (SDN) approach, testing of various load balancing techniques becomes easier and cheaper than traditional hardware-based approach. Despite the promising advantages of SDN, the novel approach is still unstable. Hence, in this experiment, performances of five different load balancing techniques-namely, random, round-robin (RR), weighted round-robin (WRR), least-connections (LC), and weighted least-connections (WLC)-were tested. The experiment was done on a single-switch topology. Mininet and POX controller were used to setup the network environment. The load balancers were also tested in two types of network conditions: with and without TCP SYN floods. After several iPerf tests, results in both network conditions indicated that RR and LC load balancers were both more than twice as fast as the one without load balancing implementation and moderately faster than random load balancer. LC and WLC were slightly faster than RR and WRR without SYN floods while RR and WRR were slightly faster with SYN floods. Future works, like testing the framework on other types of network topologies or low-level load balancing techniques, could strengthen the substantiation of stability of using SDN approach.