Network Servers Inbound Traffic Load Balancing Based on Central Decision System Results of research and model testing

Abstract

Network Load Balancing (NLB) is essential for efficient operations in distributed computer systems, especially in cloud environments. NLB systems used today usually consist of relatively simple algorithms which in certain circumstances may lead to sub-optimal balancing decisions and uneven load on computer systems that is processing user requests. Implementing improvements to load balancing algorithms by separating control and balancing parts provides a range of new possibilities for implementers. By extending control algorithm inputs so that they can become aware of the status of the entire computer system leads to far better optimization and overall performance. One of the key sources of valuable inputs is the subjective user experience that can be analysed, quantified, and used together with other inputs to create load balancing algorithms that will make better load balancing decisions. A new approach, model and algorithm is presented in this paper alongside with results of conducted tests on a prototype where in two different scenarios measured improvements superseded tenfold previous results and eliminated dropped user requests.