Deep Learning-Based Resource Utilization Prediction for Multi-pod Applications in Kubernetes

Cloud computing has become the dominant approach in computer systems due to the advancements in virtualization technology that support flexible and cost-effective solutions at almost any scale. Kubernetes is a cloud-based computing platform that enables concurrent execution of multiple tasks within isolated environments, allowing efficient hardware sharing across diverse platforms. This paper addresses the problem of predicting pod resource load. Specifically, we use deep learning techniques to predict the amount of resources (vCPU and RAM) used by every pod in a cluster. Unlike previous proposals that only consider single-container applications, we tackle the challenge of predicting resource allocation in multi-container and multi-language applications. To demonstrate the effectiveness of our methods, we use the Online Boutique microservice, an illustrative cloud-first application developed for the Google Cloud Platform. All datasets and code are available at https://github.com/JoseLuisC99/predictive-pod-resource-utilization .