Ivanisenko I. Methods of load balancing in distributed systems, taking into account the self-similar properties of input streams

The thesis proposes a new solution to the actual scientific and practical problem of developing load balancing methods that take into account the self-similarity of traffic and the use of resources of each node and the entire distributed system. The distributed load balancing model is improved, which takes into account the limitations on a given set of network characteristics for different classes of service and time variation of fractal traffic parameters. The method of calculating the imbalance of resources in a distributed system has been further developed, which includes a comprehensive measurement of the overall system imbalance level and takes into account server resource weights. Investigation of the fractal properties of additive traffic that occur when multiplexed data streams are carried out. Based on these results, a load balancing method is proposed that takes into account the multifractal properties of additive computer traffic. The use of these methods makes it possible to increase the utilization of system resources by sending heterogeneous information flows to less loaded resources.