JUCS - Journal of Universal Computer Science 32(4): 555-583, doi: 10.3897/jucs.166567
Latency-Aware Orchestration of Microservices in Heterogeneous Kubernetes Clusters Using Reinforcement Learning
Sava Stanisic‡, Borislav Djordjevic§, Branislav Belotic‡, Olga Ristic‡, Ivan Tot|, Kristina Zivanovic¶, Dimitrije Kolasinac¶‡ University of Kragujevac, Cacak, Serbia§ Mihajlo Pupin Institute, Belgrade, Serbia| University of Defence, Military Academy, Belgrade, Serbia¶ University of Belgrade, Belgrade, Serbia
Corresponding author: Sava Stanisic ( sava.stanisic@vs.rs ) © Sava Stanisic, Borislav Djordjevic, Branislav Belotic, Olga Ristic, Ivan Tot, Kristina Zivanovic, Dimitrije Kolasinac. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Citation:
Stanisic S, Djordjevic B, Belotic B, Ristic O, Tot I, Zivanovic K, Kolasinac D (2026) Latency-Aware Orchestration of Microservices in Heterogeneous Kubernetes Clusters Using Reinforcement Learning. JUCS - Journal of Universal Computer Science 32(4): 555-583. https://doi.org/10.3897/jucs.166567 |  |
Abstract
he orchestration of microservices in distributed cloud environments poses significant challenges due to the heterogeneous nature of cluster nodes and dynamic workload patterns. Traditional scheduling strategies in Kubernetes often fail to optimize latency-sensitive applications effectively. This paper proposes a latency-aware orchestration framework that integrates reinforcement learning techniques to dynamically schedule and migrate microservices across heterogeneous Kubernetes clusters. The proposed approach leverages a deep Q-network (DQN) agent trained to minimize end-to-end response times while balancing resource utilization and avoiding service-level objective (SLO) violations. Experiments conducted on a hybrid testbed comprising virtual and physical nodes demonstrate that the reinforcement learning-based scheduler reduces latency by up to 25% compared to default Kubernetes scheduling policies. The results highlight the potential of intelligent orchestration methods to enhance performance in complex cloud-native deployments.
Keywords
Kubernetes, Microservices, Reinforcement Learning, Latency Optimization, Cloud Computing, Deep Q-Network, Orchestration, Heterogeneous Clusters