AWS - Cost breakdown - Vasilis Plavos

Table of Contents

🧾 Overview of Cost Considerations by Service

💡 On-demand vs provisioned:
- Provisioned is cheaper if access is predictable. Use Auto Scaling if uncertain.
🔁 Avoid hot partitions: Distribute access evenly using good partition key design.
⏳ Enable TTL (Time To Live): Automatically delete stale data.
💾 Use DAX or caching (e.g., with Amplify or Lambda) to reduce reads.
📦 Watch storage: Monitor size growth. Large payloads = large bills.
🧹 Clean up unused tables in non-prod environments.

💡 Use only necessary Amplify features:
Hosting, auth, storage, GraphQL APIs can rack up costs—disable unused categories.
💸 Hosting costs scale with usage + storage.
Optimize asset size and invalidate old builds.
🔁 Avoid redeploying often, especially for small changes.
🔒 Use environment-specific backends, but clean up unused ones.

📋 Free itself, but can create cost-incurring resources.
🛑 Be careful with templates that create high-tier instances/services by default.
🧹 Remove old stacks/environments in test/dev.

📊 Logs and metrics cost money! Especially retention.
- Keep short log retention for dev/test (e.g., 3 days).
- Set log filters selectively (don’t log everything).
🧽 Use aws logs delete-log-group to clean up unused logs.
💡 Consider exporting logs to S3 and deleting from CW if needed long-term.

💸 REST APIs are more expensive than HTTP APIs.
- Use HTTP APIs when you don’t need advanced features (e.g. authorizers, WAF).
🧪 Use usage plans + throttling + quotas.
🔒 Enable caching to reduce Lambda invocations.

👀 No direct cost, but indirect:
- Ensure least privilege: Accidental access to high-cost resources can spike bills.
- Monitor role usage (some features are billable).

Use AWS Budgets and Cost Anomaly Detection.
Enable notifications on spikes in Lambda invocations, API calls, storage usage, etc.

Goal	Strategy
Reduce Lambda cost	Optimize memory + runtime, use async when possible
Reduce API Gateway cost	Switch to HTTP APIs, enable caching, combine endpoints
Reduce DynamoDB cost	Provisioned capacity + Auto Scaling, optimize reads/writes
Reduce Amplify cost	Host only necessary features, minimize build frequency
Reduce CloudWatch cost	Shorten log retention, remove unnecessary metrics/logs
General cleanup	Automate stack/resource teardown in dev, especially Amplify environments