Dynamic configuration of Kubernetes containers resources with SLA classes

Kubernetes is an open-source system for automating containerized applications’ deployment, scaling, and management in various industrial and academic environments. Currently, the limits of resources allocated to Kubernetes containers such as CPU, memory, and storage are statically configured. Typically, resource configuration is handled by experts who understand the resources needed by containers. However, non-expert users might struggle with this task. To address this challenge, we propose a novel approach in this paper that dynamically configures container resource limits based on predefined SLA classes. The key innovation of our approach is that it allows specifying ranges for container resource limits, instead of setting a fixed resource limit, and dynamically arranging the limits. The goal of this method is to dynamically determine the most appropriate resource limits for containers based on the current load of the Kubernetes cluster, optimizing resource allocation. For example, when the system load is high, the limits of container resources are reduced to allow more containers to run simultaneously. In contrast, resource limits can be increased when the load is low to optimize the quality of experience. Quality of experience (QoE) measures the delight or annoyance of a customer’s experience with a service. The core idea of this paper is to make it easier for non-expert users to configure resource limits, relax the rigid constraints of static resource limits, and enable the system to dynamically adjust resource allocation according to the execution context, such as activity peaks. Experiments confirm our intuition and provide insight to move forward with this idea.