Optimizing Operational Costs through Cloud Service Usage Management

Coupang manages cloud costs through a Cloud FinOps approach where Finance and Engineering work together, simultaneously pursuing budget compliance and savings targets by team. The core principles were budget allocation and compliance, target savings, and the leadership principle of Hate Waste.

First, an optimization project team centered on Cloud Infrastructure Engineers and TPMs was formed to build a collaborative structure with domain teams. This team helped each service understand the cloud's variable cost model and provided analytics environments and tools to enable decision-making based on usage data.

Cost reduction proceeded in two directions.

• Use less: In non-production environments, AWS resources were changed to auto-start only when needed, reducing costs by 25%.
• Pay less: Unused EC2 resources were identified and manually cleaned up, and the billing structure was adjusted to match usage patterns.

Infrastructure optimization was carried out in parallel. In the process of upgrading all single instances to the latest generation, AMD and ARM architectures were extensively validated, and internal products were migrated to AMD CPU-based cloud services, improving price-to-performance by 20%. For EMR, software versions were upgraded to leverage Spot Instances and the instance fleet feature was applied, cutting overall costs by 25%.

In storage, EBS and S3 received focused attention. EBS GP3 underwent performance testing with various tools, confirming that 500 to 1000 live volumes could be migrated without significant impact, and over 50PB was moved using S3 Intelligent-Tiering. However, a lesson was also learned that workloads with billions of very small objects could actually see increased costs, requiring ongoing policy adjustments including S3 lifecycle filter in such cases.

As a result, Coupang reduced AWS on-demand costs by millions of dollars in 2021. At the same time, by continuously identifying areas that need more analytics tools, monitoring, and automation, Coupang is working to establish cloud cost optimization not as a one-time effort but as an ongoing operational system.