Method of load balancing for computer cluster of data processing center

Abstract

The article presents a description of the load balancing method for a computing cluster of a data processing center (DPC), which is based on a probabilistic approach to proactive forecasting of packet traffic states, formed on the basis of the results of its statistical, nonlinear and spectral analysis. The fractal properties of network traffic are the rationale for the possibility of prediction, allow with a fairly high probability to predict the appearance of bursts and drops in its activity at certain time intervals, identify periods of possible overload of servers and network equipment, and make it possible to develop methods for effective planning and distribution of tasks within the data center, ensuring a statistically uniform loading its functional elements. The spectral analysis of the time series is carried out according to the normalized deviations of the actual levels from the smoothed ones. The absence of significant peaks in the spectral estimates indicates the absence of periodic fluctuations. It is shown that the summation of cycles of different periods of the dynamics of the time series, based on the use of the most significant harmonics of the spectrum, determines the moments of occurrence of subsequent anomalies in its development. The process of identifying significant harmonics of the spectrum is based on the study of its spectral power density using the Fourier transform. The developed method is able to provide a solution to the problem of efficient planning and distribution of tasks in a data center computing cluster in order to optimize the use of resources, speed up task execution time and reduce application processing costs.