Autoscaling with Karpenter in AWS EKS

Karpenter is an open-source Kubernetes node autoscaler designed to simplify and enhance workload scaling by dynamically provisioning nodes based on real-time demands. Unlike traditional Kubernetes cluster-autoscalers, Karpenter offers faster scaling, better resource optimization, and seamless integration with cloud provider features like Spot Instances.

The project involved implementing Karpenter Autoscaler to optimize resource utilization in AWS EKS clusters, focusing on replacing the traditional cluster-autoscaler to enhance efficiency, reduce costs, and improve scalability.

Key Achievements:

• Migrated Kubernetes Cluster Autoscaler to Karpenter Autoscaler in 16 EKS clusters with zero downtime.
• Achieved cost savings ranging from 42% to 74% per cluster by dynamically scaling workloads based on demand.
• Configured Karpenter to leverage AWS EC2 Spot Instances, balancing performance and cost-efficiency.
• Deployed Multi-Arch node pools to support diverse workloads efficiently.
• Improved node provisioning speed, ensuring seamless scaling during traffic spikes.

Example from one of 16 EKS clusters:
Below are the savings insights before and after implementing Karpenter autoscaling, achieving a remarkable 70.26% cost reduction.