Skip to content
Overview
Rose Rocket powers high-throughput transportation management software on Amazon EKS, handling a wide mix of real-time services and ephemeral jobs. This mix created substantial challenges when trying to balance performance, availability, and cost, especially as traffic patterns fluctuated throughout the day.
Despite running dynamic workloads, the team was relying on manual scaling and conservative resource allocation. Spot adoption was limited due to concerns around reliability, and rightsizing was difficult to execute consistently across such varied workloads.
By partnering with nOps, Rose Rocket introduced intelligent workload classification, automated rightsizing, and Spot-aware Karpenter tuning, resulting in major infrastructure savings and improved efficiency across their EKS environments.
The Solution
nOps helped Rose Rocket take a data-driven, automated approach to EKS optimization. Together, they:
- Automatically classified workloads by availability, duration, and volatility to guide scaling and Spot decisions
- Rightsized containers using nOps’ container rightsizing policies, with built-in buffers to handle bursty traffic and avoid throttling safely.
- Enabled intelligent Spot usage, ensuring only tolerant workloads were shifted to Spot without violating SLAs
- Tuned Karpenter to dynamically scale both persistent and ephemeral workloads with improved bin packing
- Reduced excess capacity by leveraging node-level visibility and proactive rightsizing recommendations
The Results
- 59% cost reduction across EKS through Spot adoption, rightsizing, and Karpenter tuning
- Maintained application performance and stability with increased container efficiency
- Significantly expanded Spot Instance usage in production without risk
- Reduced overprovisioning at both the container and node levels
- Gained visibility into workload-specific cost-performance tradeoffs
- Automated scaling and eliminated manual intervention
- Improved overall cluster efficiency while preserving flexibility for engineering teams
