Q-Learning-Based Lightweight Task Orchestrator: A Lightweight Q-Learning-Based Scheduler for Service Level Agreement-Aware Container Placement in Heterogeneous Clusters


== Abstract ==

<p>Efficient management of workloads in diverse container clusters requires maintaining a balance between Service Level Agreement (SLA) compliance, Quality of Service (QoS), energy efficiency, and security, despite differences in resources and architectures. This research introduces Federated Q-Learning&ndash;based Lightweight Task Orchestrator (F-Q-LiTO), a compact and intelligent orchestration framework that combines predictive modeling, approximate data structures, and security filtering to enable adaptive task placement across distributed environments. Unlike complex deep reinforcement learning models such as Deep Reinforcement Model (DeepRM) or traditional heuristic schedulers like Kubernetes BinPacking, F-Q-LiTO uses tabular Q-learning enhanced with federated aggregation, which significantly reduces computational and communication overhead, making it ideal for edge computing environments with limited resources. The framework incorporates a Long Short-Term Memory (LSTM)&ndash;based predictor for proactive resource forecasting, a Count-Min Sketch for scalable resource utilization estimation, and an XOR filter for efficient and lightweight security enforcement. Experimental results demonstrate that F-Q-LiTO achieves 98.6% task completion, 96.8% SLA satisfaction, and reduces energy consumption to 180.5 kilowatt-hours (kWh). It outperforms DeepRM and Kubernetes by achieving 34% fewer missed deadlines and up to 30% lower energy imbalance. The system converges quickly&mdash;by Episode 6&mdash;and maintains cluster fairness (Jain&rsquo;s Fairness Index= 0.98) along with priority-aware placement accuracy of 93.2%. Security analysis shows that F-Q-LiTO successfully blocks 98.5% of unauthorized task placements while using only 0.3 megabytes (MB) of memory. Overall, F-Q-LiTO demonstrates that a federated and lightweight reinforcement learning approach can deliver scalable, secure, and QoS-aware orchestration for modern edge and multi-cloud computing environments without compromising performance or efficiency.OPEN ACCESS Received: 24/07/2025 Accepted: 14/10/2025 Published: 15/12/2025</p>



== Document ==
<pdf>Media:Draft_Sanchez Pinedo_731450279-2739-document.pdf</pdf>