Implementing Cost Optimization
1. Rightsizing Service Instances
| Step | Detail |
|---|---|
| Measure | p95 CPU/mem over 7–14 days |
| Adjust | requests = p95 + headroom |
| Tools | VPA recommender, Goldilocks, Kubecost |
2. Implementing Auto-Scaling
| Aspect | Detail |
|---|---|
| Scale to zero | Knative, Cloud Run for low traffic |
| Scale on real signal | RPS, queue depth, not just CPU |
| Schedule | Off-hours scale-down for non-prod |
3. Using Spot Instances
| Aspect | Detail |
|---|---|
| Savings | 60–90% off |
| Risk | Interruption with ~2 min notice |
| Use for | Stateless, batch, fault-tolerant |
| Tools | Karpenter, Spot.io |
4. Implementing Resource Limits
| Practice | Detail |
|---|---|
| Set limits | Prevent noisy neighbor |
| ResourceQuota | Per namespace cap |
| CPU limits caveat | Throttling — often skip; keep mem limit |
5. Optimizing Database Usage
| Tactic | Detail |
|---|---|
| Right-size | IOPS + storage tier |
| Read replicas | Offload analytics |
| Aurora Serverless | Scale-to-zero (v2) |
| Archive cold | To S3 / Glacier |
6. Using Caching Effectively
| Layer | Saves |
|---|---|
| CDN | Origin egress + compute |
| App cache (Redis) | DB load |
| In-process | Network hops |
| Caveat | Watch cache size cost vs DB |
7. Implementing Data Lifecycle
| Tier | Detail |
|---|---|
| Hot | SSD; recent |
| Warm | HDD / S3 Standard-IA |
| Cold / Archive | Glacier; minutes-to-hours retrieve |
| Auto-tier | S3 Intelligent-Tiering |
8. Monitoring Resource Usage
| Metric | Use |
|---|---|
| $ per request | Unit economics |
| Idle % | Rightsizing target |
| Egress GB | Cross-AZ / cross-region cost |
| Tools | Kubecost, OpenCost, CloudHealth |
9. Implementing Reserved Capacity
| Type | Detail |
|---|---|
| Savings Plans / RI | 1–3yr commit; 30–70% off |
| Coverage target | ~70% of steady-state baseline |
| Top-up | Spot/on-demand for spikes |
10. Optimizing Network Traffic
| Tactic | Detail |
|---|---|
| Same-AZ | Avoid cross-AZ egress |
| VPC endpoints | Skip NAT for AWS services |
| Compression | gzip/zstd payloads |
| CDN | Cheaper than origin egress |
11. Using FinOps Practices
| Phase | Detail |
|---|---|
| Inform | Visibility, allocation, benchmarks |
| Optimize | Rightsize, RI, waste cleanup |
| Operate | Continuous, automated, cultural |
| Framework | FinOps Foundation |
12. Implementing Cost Allocation Tags
| Tag | Use |
|---|---|
| owner / team | Showback / chargeback |
| env | prod / stage / dev |
| service | Per microservice |
| cost-center | Finance reporting |
| Enforcement | SCP / policy denies untagged resources |