Optimization of Resource Consumption in Openstack Clusters: Algorithms for Dynamic Allocation of Computing Power

The article presents a theoretical and analytical review of algorithms for optimizing resource consumption in OpenStack clusters based on dynamic allocation of computing power. The study relies on an interdisciplinary approach that combines methods of energy management in cloud data centers, reinforcement learning algorithms, and solutions at the container and virtualization infrastructure levels. Particular attention is given to strategies that include virtual machine migration decisions using rough sets and Qlearning, task scheduling based on deep reinforcement learning in heterogeneous nodes, container service load balancing with NUMAawareness, and the integration of OpenStack and Ceph to ensure high availability and scalability. Key efficiency parameters are considered, including energy savings, balanced utilization of CPU, network, and disk resources, reduction in container deployment time, mitigation of SLA violations, and high availability performance metrics. A comparative analysis of quantitative metrics demonstrates that the combined use of approaches at different architectural levels forms the foundation for building resilient and energyefficient cloud platforms. It is shown that practical applicability requires adaptation to specific cluster environments, including threshold calibration and retraining of models on local telemetry data. Limitations are identified, such as dependency on experimental parameters and the absence of standardized methods for transferring solutions to production clusters. A conceptual model of multilevel resource management in OpenStack is proposed, where algorithms are integrated into an SLAand HAoriented control system. The article will be useful for cloud system administrators and architects, researchers in resource management, and developers of solutions aimed at improving the energy efficiency and reliability of OpenStack infrastructures.