#aws
## Reserved instances
- ### Standard RIs
- up to 72% of sales in comparison to typical on-demand
- connected to the concrete region and instance type
-
- ### Convertible RIs
- up to 54% of sales
- always can change the instance type, family, os
It works if you pay for 3 years.
## Savings Plans
### Compute Savings Plans
- up to 66% of sales
- use any of ec2
- pay per $/h
### EC2 Instance Savings Plans
- up to 72% of sales
- connected to concrete instance families in the region
- for stable workload
## Spot Instances
cons:
- just on demand but cheaper
- super for stateless
pros:
- you should be ready to terminate
- not enough count of instance types in availability zone
## **Cost Management Tools**
- **Cost Explorer** – dashboards, charts, forecasts.
- **Budgets** – set budgets and alerts.
- **Trusted Advisor** – cost, performance, and security recommendations.
- **Compute Optimizer** – recommends optimal instance types.
- **Cost Anomaly Detection** – alerts on abnormal spend.
- **Pricing Calculator** – pre-calculate service costs.
- **Service Catalog** – control access to approved resources.
## **Real Tips**
- Regularly review spend with Cost Explorer.
- Build dashboards showing top spenders (S3, EC2, Data Transfer).
- Minimize CloudWatch usage – it can get expensive fast.
- Use VPC PrivateLink where possible to reduce intra-region traffic costs.
- Scale down non-prod environments outside business hours.
- Use smaller DB instance types in non-prod.
- Change the support plan only when needed, not monthly.
- Enable Lifecycle Rules for S3, ECR, snapshots, etc.
- Avoid moving audit/access logs to Glacier – high retrieval costs later.
- Remove unused KMS keys (they add up).
- Re-evaluate instance types – maybe get more memory for the same price.
- Compare Spot prices per AZ – big differences.
- Check k8s pod requests/limits – overprovisioning waste nodes.
- Cost optimization is never done — clients will always ask for more savings.
## **Additional Real Tips from Community**
- **Use Graviton (ARM)** instances where possible — up to 60% savings.
- **Migrating from gp2 to gp3** volumes — saves 15–20% on EBS.
- **Reduce inter-AZ traffic** — can save $20–30/day.
- **Avoid magnetic volumes** — unpredictable latency, not worth the cost.
- **Delete/Archive old EBS snapshots** — they cost as much as the volumes
- **Use scheduled deletion for KMS keys** and monitor usage to avoid surprises
- **Beware of PendingDeletion keys reactivation charges** — can cost thousands.
- **Hybrid cloud setup (on-prem + cloud)** can be more cost-efficient at scale.
- **Switch to IPv6** where possible.
- **Replace CloudFront + WAF + Route53** with Cloudflare if it meets your needs.
- **Use HTTP Gateway instead of API Gateway** when possible — cheaper.
- **Scale down staging/test environments after hours** (e.g. via cron + automation).
- **Use ephemeral environments** per feature/branch — destroy after use.
- **Integrate preview environments** that spin up only when needed.
- **Use Lambda + S3 + CloudFront + SQS** instead of EC2 where applicable.
- **Involve devs in infra planning** — optimize architecture early.
- **Don’t Dockerize everything** — static assets can go to S3.
- **Serverless can be free** (for small scripts within the Free Tier).
- **Avoid vendor lock-in** unless you really benefit from managed services.
- **Don’t trust AWS blindly** — always double-check their cost suggestions.
- **Spot ≠ linear saving** — it’s opportunistic access to spare capacity
- **Carpenter handles Spot interruptions well** — use it over manual config.
- **Monitor Total forecasted cost for early anomalies.**