Cost Management#

AWS bills can grow faster than your infrastructure if unmanaged. Cost optimisation is a continuous practice — not a one-time cleanup. This section covers the tools, patterns, and organisational practices for keeping AWS spend under control.

Understanding Your Bill#

AWS Cost Structure#

Common Cost Surprises#

AWS Cost Tools#

Cost Explorer#

Visualise and analyse spend over time:

• Filter by service, account, tag, region, instance type
• Group by any dimension
• Forecast future spend based on trends
• Identify cost anomalies

AWS Budgets#

Set spending thresholds and get alerts:

Budget: $5,000/month for production account
Alert 1: 80% threshold ($4,000) → email team
Alert 2: 100% threshold ($5,000) → email + SNS → Lambda auto-response
Alert 3: Forecast exceeds budget → early warning

Cost and Usage Report (CUR)#

The most detailed billing data — line-item level, exported to S3. Use with Athena for custom analysis:

SELECT line_item_product_code, SUM(line_item_unblended_cost) as cost
FROM cur_data
WHERE month = '2024-01'
GROUP BY line_item_product_code
ORDER BY cost DESC
LIMIT 10;

Pricing Models#

On-Demand vs Reserved vs Savings Plans vs Spot#

Savings Plans vs Reserved Instances#

Right-Sizing#

Match instance sizes to actual utilisation:

Current: m5.2xlarge (8 vCPU, 32 GB) — avg CPU 15%, avg memory 20%
Right-sized: m5.large (2 vCPU, 8 GB) — saves ~75%

Tools#

• AWS Compute Optimizer — analyses CloudWatch metrics, recommends instance types
• Cost Explorer Right-Sizing Recommendations — identifies underutilised instances
• Trusted Advisor — checks for idle resources, underutilised instances

Auto-Scaling#

Don’t run peak capacity 24/7:

Baseline: 2 instances (handles normal traffic)
Peak: 10 instances (auto-scales for traffic spikes)
Schedule: scale down to 1 instance at night (non-production)

Tagging Strategy#

Tags are the foundation of cost allocation:

Enforce tags with AWS Organizations SCPs or AWS Config rules. Untagged resources are unaccountable.

FinOps Practices#

The FinOps Loop#

Inform → Optimise → Operate → Repeat

1. Inform: Visibility into who spends what (tags, CUR, dashboards)
2. Optimise: Right-size, purchase commitments, eliminate waste
3. Operate: Continuous governance (budgets, alerts, anomaly detection)

Quick Wins Checklist#

• Delete unattached EBS volumes and old snapshots
• Set CloudWatch Logs retention (not “Never expire”)
• Stop dev/staging RDS instances outside business hours
• Use S3 Intelligent-Tiering for infrequently accessed data
• Enable S3 Lifecycle rules (move to Glacier after 90 days)
• Use VPC endpoints instead of NAT Gateway for AWS service traffic
• Review and terminate unused Elastic IPs, load balancers, NAT gateways
• Buy Savings Plans for steady-state compute
• Enable Cost Anomaly Detection in Cost Explorer
• Tag every resource — enforce via SCP or Config rule

Key Takeaways#

• Cost surprises come from NAT Gateway data fees, idle databases, old snapshots, and CloudWatch Logs retention.
• Savings Plans offer 30-72% discounts with more flexibility than Reserved Instances. Prefer them for compute.
• Right-sizing is the highest-impact single action — most instances run at 10-20% utilisation.
• Tags are mandatory for cost accountability. Untagged resources are invisible to cost analysis.
• FinOps is a continuous loop: inform, optimise, operate. Set budgets and alerts from day one.
• Use Cost Explorer for analysis, Budgets for alerts, Compute Optimizer for right-sizing, and CUR for deep dives.