AWS Cost Management
Price exercises use hypothetical us-east-1 (US East, N. Virginia) rates for study on 2026-10-12, not current tariffs, unless explicitly identified as a dated published fee.
Cost management is an architectural practice: define ownership, measure usage, remove waste, choose suitable pricing and verify the result. This hub is the canonical cost-learning path; the concept view provides concise exam revision.
Learning path
- Pricing models — purchasing choices, account-cohort Free Tier rules and unit-aware comparisons.
- Allocation tags and Organizations strategy — ownership, sharing and chargeback.
- Cost Explorer and Billing exports — delayed billing analysis, metrics and schemas.
- Budgets and Anomaly Detection — fixed objectives versus unexpected changes.
- EC2 optimization — rightsizing and evidence before commitments.
- Reserved Instances, Savings Plans and Commitment management — eligibility, utilization and coverage.
- Safe automation — inventory, owner approval, bounded changes and measured savings.
Service investigations
- S3 — lifecycle, retrieval, replication and access patterns.
- EBS — provisioned storage/performance and safe retention.
- EFS — availability, lifecycle and throughput modes.
- RDS and Aurora — database performance, topology and serverless/commitment choices.
- DynamoDB — item/consistency units, indexes and cache economics.
- Networking — traffic-path modeling and service-specific exemptions.
- Lambda — GB-seconds, requests, provisioned concurrency and surrounding services.
Common billing traps
Stopping EC2 leaves EBS charges running; detaching EBS does not remove its bill. Lambda VPC ENIs do not have the invented NAT hourly fee. Private VPC-local database traffic does not inherently traverse NAT. Budget/anomaly alerts and CloudWatch billing data are delayed, not real-time hard caps.
Free Tier rules depend on account creation cohort and plan. Not all S3 egress is charged, and not all endpoints are free. Sharing a reservation discount is different from sharing capacity. Savings Plans commit discounted-rate dollars per hour, not a fraction of a monthly invoice.
How to use worked examples
All illustrative rates are assumptions, not current quotes. Keep quantity, unit, Region, term, duration, price and exclusions visible. Include storage, transfer, logs, commitments and related services. Proposed drills are not reports of completed deployments.
Service-by-service cost review
| Layer | First evidence to collect | Common residual cost |
|---|---|---|
| Compute | Running hours, peak utilization and scaling | Idle commitments, EBS and IPv4 |
| Serverless | Memory/duration/request mix and concurrency | NAT, logs, API, queues and proxies |
| Object storage | Class, age, size, versions and retrieval | Early deletion, replication and requests |
| Block/file storage | Provisioned size/performance and access | Snapshots, throughput and colder-tier access |
| Databases | Queries, capacity, topology and consistency | Replicas, backup, I/O and licenses |
| Networking | Directed traffic paths and bytes | NAT/TGW/endpoint hours and processing |
| Governance | Owner/account/tag completeness | Shared/unallocated subscriptions and support |
The detailed notes retain practical comparisons and proposed labs; use the table to select an investigation, not as a universal savings checklist.
Reconciled platform estimate
For a hypothetical us-east-1 planning exercise reviewed on 2026-10-12, assume ten workers run 720 hours at $0.10/hour, 2,000 GB object storage at $0.02/GB-month, and the following separately estimated components:
| Component | Calculation or assumption | Monthly amount |
|---|---|---|
| Workers | 10 × 720 × $0.10 | $720 |
| Object storage | 2,000 × $0.02 | $40 |
| Database | Hypothetical configuration subtotal | $180 |
| Network/load balancing | Hypothetical measured-path estimate | $90 |
| Block storage/backups | Hypothetical provisioned/retained data | $70 |
| Monitoring/logging | Hypothetical ingestion/retention | $30 |
| Total | Sum of the components | $1,130 |
These are assumed inputs, not current product tariffs. License, support, taxes, credits and commitments are excluded. A useful production estimate must replace every assumption with a selected service configuration/rate and include omitted components.
If rightsizing halves worker cost, gross reduction is $360, or 31.86% of the whole estimate, not “50% off the AWS bill.” If an unused commitment still costs $250, realized savings can be only $110 before other changes.
Practical FinOps cycle
Allocate → baseline → inspect → design → approve → change → verify. The owner validates SLO/recovery requirements; finance validates the accounting view; platform/security validate access and operational safeguards. Record the estimate, measurement window and counterfactual baseline before claiming realized savings.
Use the task lessons organizational visibility, new-solution economics and existing-system improvement to practice different decision contexts.
References
Across the wiki
- EKS Cost Optimization — Kubernetes cost (Kubernetes)
- AKS FinOps, Cost Allocation, and Cloud Spend Optimization — Kubernetes cost (Azure)
- GKE Cost Optimization, FinOps, and GKE Cost Allocation Architecture — Kubernetes cost (GCP)
- Cost Optimization — Kubernetes cost (Kubernetes)