Mathematical modeling and external load-balancing methods in server clusters
Abstract
Mathematical modeling and external load-balancing methods in server clusters
Incoming article date: 03.06.2025This paper presents the design and experimental validation of an external load-balancing mechanism for server clusters that support a distributed educational network. A hybrid strategy is proposed that merges classical policies (Round Robin, Least Connections) with an evolutionary search based on a genetic algorithm. At the modeling level the user-session assignment problem is formulated as a minimization of the maximum node load under latency constraints. The solution is implemented entirely on a domestic technology stack— “1C:Enterprise” server clusters, Docker containers, and the “1C:Bus” integration middleware. Experimental results show that the new scheduling logic improves system resilience under traffic fluctuations, lowers user response times, and utilizes spare resources more efficiently, while imposing no substantial overhead on the control nodes. The study confirms the practical viability of evolutionary approaches for real-time load balancing.
Keywords: load balancing, server clusters, genetic algorithm, simulation modeling, 1C:Bus middleware