UitbreidenOS Cloud Cost Optimization
When to activate
When conducting cost audits, planning reserved instance purchases, analyzing rightsizing opportunities, or designing cost-efficient cloud architectures.
When NOT to use
For one-off cost checks or when cost is not a primary constraint of the architecture.
Instructions
Cost optimization requires systematic analysis, not guessing. Follow this framework:
1. Cost Audit Methodology
Analyze costs in 4 dimensions:
a) Compute (50% of spend typically)
- Instance type, size, utilization
- Reserved instances (RI) / Savings Plans (SP) coverage
- Spot instances for interruptible workloads
- Idle or underutilized instances
b) Storage (20% typically)
- Object storage (S3, Blob, GCS) usage and lifecycle
- Database storage (RDS, Cosmos, Firestore)
- Backup and snapshot retention
- Redundancy and replication
c) Data Transfer (20% typically)
- Inter-region transfer costs
- CDN vs. direct download tradeoffs
- NAT Gateway charges
- CloudFront, Azure CDN, Cloud CDN usage
d) Managed Services (10% typically)
- Lambda, Functions invocations and duration
- Database managed service overhead
- Message queue costs (SQS, Pub/Sub)
- Monitoring and logging (CloudWatch, Monitor, Logging)
2. Reserved Instance (RI) Analysis
Framework:
- Identify baseline compute — What instances run 24/7, year-round?
- Calculate RI payback — If I commit for 1 year, when do I break even?
- Right-size before buying — Ensure instance is right size; RI for undersized instance is waste
- Blend RI + On-Demand — Use RIs for baseline, On-Demand for spikes
- Track RI utilization — Aim for >80% to justify the commitment
RI ROI Calculation:
On-Demand annual cost: 500 instances × $730/year × $1 = $365,000
1-year RI (40% discount): 500 instances × $730/year × 0.6 = $219,000
Payback period: $219,000 ÷ ($365,000/12) = 7.2 months
Decision: Buy 1-year RIs if confidence high on compute needs
Decision: Buy 3-year RIs if needs are very stable (30-50% additional discount)
RI Purchasing Strategy (AWS):
| Workload Type | Recommendation | Rationale |
|---|---|---|
| Production baseline (24/7) | 3-year RI | 50% discount, locked commitment |
| Semi-production (16h/day) | 1-year RI | Balances discount vs. flexibility |
| Ephemeral/testing | Spot instances | 70% discount, interruption acceptable |
| Variable peak load | On-Demand | No commitment, pay as you go |
Savings Plans (more flexible):
- Compute Savings Plans: 1-3 year commitment, discount applies across instance families
- Example: "I commit to $2000/month compute spend" applies across t3, m5, c5, etc.
3. Spot Instance Strategy
Spot instances cost 70% less but can be interrupted. Use for:
- Batch jobs with checkpointing
- Stateless application servers behind autoscaling
- Data processing (Spark, Hadoop)
- CI/CD pipeline runners
- Development/testing
Do NOT use for:
- Databases requiring data persistence
- Single-instance applications
- Real-time critical services
Implementation:
Primary fleet: 3× on-demand m5.xlarge + 2× spot m5.xlarge
Target group weight: On-Demand=60%, Spot=40%
Fallback: If spot unavailable, autoscale on-demand
Expected savings: 40% × 70% = 28% reduction
4. Tagging Strategy
Mandatory tags for cost allocation:
Project: [project-name]
Environment: [dev/staging/prod]
Owner: [team/person]
CostCenter: [cost-center-code]
Application: [app-name]
ManagedBy: [terraform/manual/cloudformation]
Query costs by tag:
# AWS
aws ce get-cost-and-usage \
--time-period StartDate=2026-01-01,EndDate=2026-06-30 \
--granularity MONTHLY \
--metrics "UnblendedCost" \
--group-by Type=DIMENSION,Key=TAG_Environment
# Result: Dev costs separate from Prod
5. Storage Lifecycle Policies
S3/Blob Lifecycle:
Current (0-30 days): Standard storage ($0.023/GB)
Archive (31-90 days): Infrequent access ($0.0125/GB)
Cold (91-365 days): Glacier ($0.004/GB)
Delete (>365 days): Remove
Cost impact: 100 TB data
- All standard: $2,300/month
- Tiered lifecycle: $800/month first month, then $400/month
- Annual savings: $19,200
Implementation:
{
"Rules": [
{
"Id": "transition-to-ia",
"Filter": {"Prefix": "logs/"},
"Transitions": [
{
"Days": 30,
"StorageClass": "STANDARD_IA"
},
{
"Days": 90,
"StorageClass": "GLACIER"
}
],
"Expiration": {"Days": 365}
}
]
}
6. Compute Rightsizing
Process:
- Identify underutilized instances — CloudWatch metrics: CPU <20%, memory <40%
- Right-size before commitment — Move t3.xlarge to t3.medium before buying RI
- Test smaller size — Stage change, validate performance
- Measure savings — Track price difference
Example:
Current: m5.2xlarge (8 vCPU, 32 GB RAM) = $0.384/hour
Analysis: Peak CPU = 35%, peak memory = 28%
New: m5.xlarge (4 vCPU, 16 GB RAM) = $0.192/hour
Savings: $0.192/hour × 730 hours/month = $140/month ($1,680/year)
Risk: Validate under peak load before fully committing
7. Data Transfer Costs
Inter-region transfer: $0.02/GB expensive. Minimize by:
- Using same region when possible
- Caching (CloudFront, CDN)
- Replicating only critical data
Example cost:
Scenario A: Single region + read from other regions
- Daily read: 1 TB from us-west-2 to eu-west-1
- Cost: 30 TB × $0.02/GB = $600/month
Scenario B: Multi-region read replicas
- Database replication: $200/month (RDS)
- Local reads: $0 data transfer
- Total: $200/month
- Savings: $400/month ($4,800/year)
8. Managed Service Optimization
Lambda cost optimization:
- Analyze execution time distribution
- Batch invocations to reduce function calls
- Increase memory (faster execution, reduced duration cost)
- Use reserved capacity for predictable baseline
RDS cost optimization:
- Right-size instance type
- Use read replicas instead of scaling vertically
- Enable automated backups (cheaper than manual)
- Use Aurora for variable workloads (pay per request)
Cache optimization:
- ElastiCache for hot data vs. querying database repeatedly
- CloudFront for static assets
- TTL tuning: balance freshness vs. requests
9. Cost Monitoring Dashboard
Create dashboard tracking:
Metric Target Frequency
Daily spend $X ±10% Daily
RI utilization >80% Weekly
Top 5 cost drivers [List] Weekly
Cost per request $[X] ±5% Weekly
Reserved vs. On-Demand [Ratio] Monthly
Month-to-date vs budget <100% Daily
10. Cost Governance
Monthly review process:
- Forecast — Project end-of-month spend
- Analyze variance — Why is spend up/down vs. projection?
- Recommend — RIs to buy, instances to rightsize, savings to implement
- Track — Measure actual savings vs. projected
- Report — Share with finance and engineering leadership
Example
Cost Optimization Project: SaaS Platform
Baseline Analysis:
Month 1 Spend: $85,000
- Compute: $42,500 (50%)
- Storage: $17,000 (20%)
- Data transfer: $17,000 (20%)
- Managed services: $8,500 (10%)
Optimization Plan:
Reserved Instances ($42,500 → $25,500)
- Identify baseline compute: 200× t3.xlarge (24/7)
- Buy 1-year RIs at 40% discount
- Savings: $17,000/month
Spot Instances ($8,500 → $3,000)
- Move ephemeral jobs to spot (70% discount)
- Maintain availability with fallback to on-demand
- Savings: $5,500/month
Storage Lifecycle ($17,000 → $8,500)
- Implement S3 lifecycle: Standard → IA → Glacier
- Archive logs older than 90 days
- Savings: $8,500/month
Data Transfer Reduction ($17,000 → $12,000)
- Enable CloudFront for static assets
- Migrate cross-region reads to read replicas
- Savings: $5,000/month
Total Projected Savings:
- $17,000 (RI) + $5,500 (spot) + $8,500 (storage) + $5,000 (transfer) = $36,000/month
- Baseline: $85,000 → Optimized: $49,000 (42% reduction)
Implementation Timeline:
- Week 1: Buy 1-year RIs for identified baseline
- Week 2: Implement S3 lifecycle policies
- Week 3: Configure CloudFront for static assets
- Week 4: Migrate batch jobs to spot instances
- Week 5: Monitor and adjust
Tracking:
Month 2 (post-optimization):
- Target: $49,000 (42% reduction)
- Actual: $51,200 (actual reduction: 40%)
- Variance: +$2,200 (within 5% tolerance)
- Next month actions: Fine-tune RI allocation, test spot interruption handling
Version: 1.0
Last updated: 2026-06-15