METHODS FOR CONTROLLING HARDWARE RESOURCE USAGE BY AUTONOMOUS SOFTWARE AGENTS IN MULTI-TENANT CLUSTER ENVIRONMENTS: REVIEW OF APPROACHES AND GAP ANALYSIS

This study presents a systematic review of methods for controlling hardware resource usage by autonomous software agents in multi-tenant cluster environments, based on 25 sources (2020–2026) selected via the Kitchenham methodology across six search rounds in Google Scholar, arXiv, USPTO, Espacenet, IEEE Xplore, and ACM Digital Library. A chronology of resource control technologies (2007–2026) is presented, from cgroup v1/v2 to eBPF subsystem evolution, agent architectures, and EU regulatory requirements. Four categories of approaches are examined: cgroup v2 limits, application-layer network policies, eBPF instrumentation at the OS kernel level, and cloud platform APIs. A detailed technical analysis of seven eBPF components is provided, being the category closest to solving the problem. A comparative analysis of 14 systems across five criteria (Ring-0 enforcement, zero-gap update, proactive verification, cryptographic attestation, and control-plane protection) against NIS2, DORA, and EU AI Act requirements reveals three structural gaps: absence of zero-gap configuration updates in Ring-0 systems for autonomous agents, absence of a proactive oracle interface for agent action verification, and absence of architectures combining Ring-0 enforcement with a cryptographically verifiable audit chain. The identified gaps define the directions for further research.