AWS cost optimization checklist

AWS bills grow quietly — oversized RDS instances, NAT Gateway traffic, orphaned EBS volumes and dev environments left running overnight. Cost optimization is not a one-time exercise; it is a recurring practice tied to metrics, tagging and ownership. Use this checklist after migration or quarterly for production estates Kode Builder operates or advises on.

1. Visibility, Tagging and Budgets

• Cost allocation tags applied (environment, service, team, cost-center)
• AWS Budgets configured with email/Slack alerts at 80% and 100%
• Cost Explorer saved reports per product line reviewed monthly
• Anomaly detection enabled for unexpected spikes
• Separate accounts or OU structure for prod vs non-prod

2. Compute Right-Sizing (EC2, ECS, EKS, Lambda)

• CloudWatch CPU/memory utilization reviewed over 14–30 days
• EC2 instances below 20% avg CPU considered for downgrade or Graviton
• ECS/Fargate task CPU and memory right-sized from actual metrics
• EKS node groups use cluster autoscaler; remove over-provisioned pools
• Lambda memory tuned via power tuning; remove unused functions
• Dev/staging auto-shutdown schedules or smaller instance classes

3. Reserved Capacity and Savings Plans

• Steady-state baseline identified (not peak) before buying RIs/SPs
• Compute Savings Plans or EC2 Instance Savings Plans evaluated
• RDS Reserved Instances for production databases with stable sizing
• ElastiCache reservations for persistent cache clusters
• Commitment coverage reviewed quarterly as traffic changes

4. Storage and Data Lifecycle

• Unattached EBS volumes and old snapshots identified and deleted
• S3 lifecycle policies move infrequent data to IA/Glacier tiers
• S3 Intelligent-Tiering or storage class analysis enabled where appropriate
• RDS storage autoscaling limits set; old manual snapshots archived or removed
• CloudWatch log retention limits configured (avoid infinite log storage)

5. Networking and Data Transfer

• NAT Gateway costs reviewed — VPC endpoints for S3/DynamoDB reduce egress
• Cross-AZ traffic minimized where architecture allows
• CloudFront used for static assets and cacheable API responses
• Inter-region replication justified; unnecessary duplication removed

6. Databases and Caching

• RDS instance class matches workload (Aurora Serverless v2 for variable load)
• Read replicas justified by query load, not speculatively provisioned
• ElastiCache cluster size validated; TTLs prevent unbounded memory growth
• Idle database connections and N+1 query patterns addressed in application layer

7. Ongoing FinOps Process

• Monthly cost review meeting with engineering and finance
• New services require cost estimate in design doc
• Trusted Advisor or Cost Optimization Hub recommendations triaged
• Post-incident reviews include cost impact of scaling decisions