Algorithm for Load Balancing of a Data Processing Center Based on a Nonlinear Forecast Model

The purpose of the article is to increase the efficiency of the cloud data center load distribution and balancing system by developing and applying a mechanism for predicting network traffic conditions characterized by fractal self-similarity. When developing the predictive model, methods of nonlinear dynamics were used, taking into account the statistical self-similarity of the load and providing a solution to the problem of predicting the moments of its expected bursts. The randomness of network traffic was checked by calculating the spectrum of Lyapunov exponents. The mathematical model and the dynamic algorithm of the predictive model of the state of a nonlinear system are presented in the form of a system of discrete mappings of previous and subsequent values of a time series and a regression approximating polynomial connecting them. The presented algorithm differs from the existing ones by taking into account the features of fractal self-similarity of the input load, which negatively affects quality indicators, using a predictive model developed on the basis of nonlinear dynamics methods, as well as the possibility of choosing rational balancing parameters according to the criterion of uniform loading of server resources. The article shows that the dynamic load balancing algorithm, based on nonlinear approaches and predictive models, allows solving load distribution problems between servers of data center clusters more efficiently than traditional methods. The conclusion about the effectiveness of the developed algorithm is substantiated.