IMPROVING CLOUD SYSTEM PERFORMANCE THROUGH ADAPTIVE RESOURCE OPTIMIZATION BASED ON GENETIC ALGORITHMS

Abstract

The article investigates an approach to improving the performance of cloud systems through adaptive resource optimization based on genetic algorithms (GA). Particular attention is paid to evaluating system efficiency under high-load conditions in a hybrid AWS cloud environment that simulates real-world usage scenarios. The study was conducted on an architecture comprising three t3.small EC2 instances, which acted as request processing servers, and one t3.medium EC2 instance that served as a router. The router hosted genetic algorithms (GA) and a neural network (NN) that predicted peak loads and helped adaptively distribute requests.

The research methodology is based on load testing using the Gatling tool, which enables user behavior simulation and system performance analysis under various load conditions. Key performance parameters such as total execution time, resource usage cost, and actual CPU and memory utilization were analyzed. A series of experiments was conducted with various configurations, including the use of the Classic Genetic Algorithm (Classic GA), the Multi-Objective Genetic Algorithm (Multi-Objective GA), and the Hybrid GA + RL algorithm with a neural network trained for 15 minutes, 30 minutes, 1 hour, and 12 hours.

The results demonstrated that using genetic algorithms significantly improves system performance compared to traditional load balancing approaches. The Hybrid GA + RL approach with 12 hours of neural network training proved to be the most effective, achieving the lowest execution time, optimal CPU and memory usage, and minimal resource costs among all tested configurations. The Multi-Objective GA also outperformed the classic algorithm, particularly in cases of unstable workloads.

Thus, the obtained results confirm the feasibility of applying adaptive optimization based on genetic algorithms and neural networks in AWS cloud systems. The proposed approaches provide enhanced performance, cost reduction, and improved system stability. The findings can be useful for engineers working with cloud services as well as developers of scalable, high-load web applications.