Migrate to Amazon EKS: Data plane cost modeling with Karpenter and KWOK

Migrate to Amazon EKS: Data plane cost modeling with Karpenter and KWOK Solution overview Solution walkthrough Prerequisites Step 1: Create the source EKS cluster Step 2: Deploy an example workload Step 3: Extract cluster configuration with Velero Step 4: Create the destination EKS cluster Step 5: Deploy Karpenter with the KWOK provider Step 6: Restore the backup Clean up Conclusion About the authors When migrating Kubernetes clusters to Amazon Elastic Kubernetes Service (Amazon EKS) , organizations typically follow three phases: assessment, mobilize, and migrate and modernize. The assessment phase involves evaluating technical feasibility for Amazon EKS workloads, analyzing current Kubernetes environments, identifying compatibility issues, estimating costs, and determining timelines with business impact considerations. During the mobilize phase, organizations create detailed migration plans, establish EKS environments with proper networking and security, train teams, and develop testing procedures. The final migrate and modernize phase involves transferring applications and data, validating functionality, implementing cloud-centered features, optimizing resources and costs, and enhancing observability to fully use AWS capabilities. One of the most significant challenges organizations face during the process is cost estimation, which happens in the assessment phase. Karpenter is an open source Kubernetes node autoscaler that efficiently provisions just-in-time compute resources to match workload demands. Unlike traditional autoscalers, Karpenter directly integrates with cloud providers to make intelligent, real-time decisions about instance types, availability zones, and capacity options. It evaluates pod requirements and constraints to select optimal instances, considering factors such as CPU, memory, price, and availability. Karpenter can consolidate workloads for cost efficiency and rapidly scale from zero to handle sudden demand spikes. It supports both spot and on-demand instances, and automatically terminates nodes when they’re no longer needed, optimizing cluster resource utilization and reducing cloud costs. Karpenter uses the concept of Providers to interact with different infrastructure platforms for provisioning and managing compute resources. KWOK (Kubernetes WithOut Kubelet) is a toolkit that simulates data plane nodes without allocating actual infrastructure, and can be used as a provider to create lightweight testing environments that enable developers to validate provisioning decisions, try various (virtual) instance types, and debug scaling behaviors.