FSCLB: A dynamic feedback-based server cluster load balancing mechanism for data centers over Software Defined Networking

Abstract Data centers leverage specific load balancers to achieve a load equilibrium for server clusters. Due to larger occupancy on system resources and the simplicity of load balancing algorithms, these load balancers cannot obtain each server’s real loads and dynamic load variation between servers precisely under some situations so that data centers’ service levels are lowered. Although researchers have put forward a lot of solutions, these solutions often don't take full use of high connectivity of data centers and depend on limited traffic distribution features, resulting in a poor utilization of network resources. There is a need to adopt some innovative techniques to improve shortcomings in these solutions and realize high-level load balancing between servers. For this purpose, we exploit the technical superiorities of SDN and present a dynamic feedback-based server cluster load balancing mechanism. We carry out the research from request allocation and traffic scheduling. To obtain a server’s accurate load statistics, we design a self-adaptive server data acquisition method to select a best performance server for responding new requests. For request forwarding, we focus on dealing with the mixed flows in the network. We classify elephant flows using a two-stage elephant flow identification algorithm and use a flow classification-based route algorithm to forward elephant flows and mice flows. The experimental results show that compared with other algorithms, our load balancing mechanism can efficiently reduce the overall loads of the server cluster, improve network throughput and shorten the finish time and average delays of flows.