International Journal of Sustainability and Innovation in Engineering (IJSIE)
2024
https://www.doi.org/10.56830/IJSIE202405
Author
Sandeep Reddy Gundla
Abstract
This study investigates Kubernetes scheduling that is optimized with AI through Node Affinity to maximize the deployment and performance of Java microservices. It also reviews the drawbacks of conventional, rule-based schedulers —the lack of adaptability in workloads and their resource requirements that tend to cause inefficient node assignment, resource partitioning, and performance issues. Given that the impending integration of machine learning algorithms into the core Kubernetes codebase will support only supervised learning of the resource requirements and reinforcement learning of adaptive schedules, the proposed framework will enhance the native scheduler in Kubernetes with the ability to make data-derived decisions. A custom scheduler plugin makes all of them use historical and real-time scale data as well as metrics, such as CPU, memory, I/O, pod latency, throughput, predictive node scoring, and affinity-based pod placement. The experiments conducted in an isolated Kubernetes cluster to test the AI-optimized scheduling reveal a 12% decrease in the mean absolute error when forecasting the required resources, a 25% throughput improvement in microservices, an 18% increase in CPU and memory utilization, and a 15% decline in response time as compared to the default scheduler. The placement efficiency of the Java microservices also increases by 22%, which can affirm successful matching of the microservice specifications to the node capabilities through the framework. These findings indicate high performance, scalability, and cost-effectiveness and provide recommendations to help industries incorporate AI models into production Kubernetes practices. Work in the future will include deep learning advanced architectures, continuous ad-hoc model retraining, and the extension to heterogeneous workloads in the cloud. This framework eliminates the manual overhead of configuration. It can be continuously optimized, shifting the need for resilient and efficient operations as part of the larger-scale Kubernetes cluster and the ability to scale economically in any location around the globe. This research affirms the power of AI-based scheduling, which can drive the container orchestration industry that focuses on optimizing object or node affinity-based selection decisions, as seen in Java microservices.
Keywords;
AI-Optimized Scheduling, Kubernetes, Java Microservices, Node Affinity, Reinforcement Learning
Download Full Article 