Multi-dimensional resource placement algorithm based on parallel genetic algorithm

Abstract

With the advancement of cloud-native technologies, container cluster management systems such as Kubernetes, Swarm, and Mesos have emerged. Due to its superior container orchestration capabilities, Kubernetes has been widely adopted across diverse domains and is now the industry-preferred solution for container cluster management. However, Kubernetes primarily relies on a single resource dimension for Pod placement, which often leads to imbalanced resource utilization and single-resource bottlenecks. To address this limitation, we optimize the Pod placement strategy in Kubernetes by designing a parallel genetic algorithm based on the island model, which accounts for multi-dimensional resource consumption in cloud-native environments. The genetic algorithm is tailored to the cloud-native context through enhancements in genetic coding design, initial population generation, and objective function formulation. By integrating the island model with genetic algorithms, our parallel optimization approach improves computational efficiency and addresses the NP-hard challenge of resource placement in cloud environments. Experimental results demonstrate that the proposed algorithm reduces the average single-prediction time by 42.5 %, achieves a cluster resource utilization rate of 93.77 %, and attains a parallel speedup ratio of 3.681. Furthermore, it mitigates resource imbalance and enhances utilization efficiency across clusters of varying scales.