Goal-Driven Autonomous Agents for SLA-Aware Network Orchestration

Abstract

Achieving rigorous latency SLAs in dynamic telecommunications environments necessitates continuous optimization of network topology and capacity. This paper presents NetAgent-SLA, a goal-driven autonomous agent framework designed to monitor real-time topological and QoS metrics, implement reinforcement learning policies, and initiate SDN and cloud-API reconfigurations to maintain SLA performance. NetAgent-SLA was deployed on a multi-cloud testbed comprising on-premises SDN, Azure, and AWS platforms, demonstrating the following results:

• Maintained 99.5% SLA compliance across request rate fluctuations ranging from 500 to 2,000 per second
• Achieved a 28% reduction in 95^th percentile latency compared to static orchestration approaches
• Adapted to new optimal configurations within 30 seconds following topology changes
• Incurred less than 1.2 ms decision-loop overhead per cycle

This work provides a comprehensive explanation of system architecture, agent design, training methodology, performance evaluation, and addresses integration challenges as well as future directions for autonomous network operations.