40% AWS EKS Operating Cost Cut with NetApp Spot, Achieving Zero-Downtime Migration with SLA Guarantee

Background of Cost Optimization and Solution Adoption

The chatbot service (RB Dialog), which had migrated from on-premise to the AWS cloud, required optimization due to operating costs higher than expected. On the recommendation of the internal cloud engineering team, the company adopted the Spot by NetApp solution and successfully reduced costs. Regular Spot instances carry the risk of termination, making them difficult to apply to services requiring stability, but Spot by NetApp solved this by guaranteeing an SLA (Service Level Agreement).

How Spot by NetApp Works

This solution predicts instance availability and interruption timing through machine learning-based analysis of the Spot market. In preparation for predicted interruptions, it performs zero-downtime replacement of existing instances with new ones, and switches to On-Demand instances when supply is unavailable, preventing service disruption. For the author's team, the average Spot instance discount rate was around 60%, ultimately reducing overall cloud operating costs by about 40%.

Key Configurations for EKS Environment Application

The following configurations in the Kubernetes (EKS) environment were essential for uninterrupted operation.

• Pod Lifecycle Management: Configured startupProbe to route traffic only after the application is ready, and adjusted preStop hook and terminationGracePeriodSeconds to prevent traffic loss during termination.
• PDB (PodDisruptionBudget): Set a minimum number of pods to maintain in order to prevent simultaneous pod termination during node replacement.
• Distributed Placement Strategy: Used topologySpreadConstraints to distribute pods across various nodes and zones, preventing concentration on specific nodes and increasing availability.
• Headroom Configuration: Leveraged Spot by NetApp's Headroom feature to pre-reserve spare capacity within nodes, enabling pods to run quickly during node replacement without waiting for new node provisioning.

Validation and Results

The team used the nGrinder load testing tool to verify whether traffic was properly handled during node replacement and pod movement. By applying appropriate delay settings and a distributed placement strategy, most error cases could be prevented. This case demonstrates that Spot instances can be utilized even for services where stability is critical, proving the applicability of NetApp Spot in cloud cost optimization.