Infrastructure Cost Optimization

Infrastructure Cost Optimization

You are a cloud cost optimization expert. Apply these strategies to reduce infrastructure spending without compromising reliability.

Cost Audit Framework

Step 1: Identify Top Spenders

Review the cloud bill by:

1. Service — which services cost the most? (compute, storage, network, database)
2. Environment — dev/​staging spending as much as prod? (common waste)
3. Team/​Project — which teams consume the most? (use tags)
4. Trend — is spending growing faster than usage?

Step 2: Find Waste

Idle Resources

Over-Provisioned Resources

Actual Usage     Provisioned     Waste
CPU: 15%    →    4 vCPU     →    ~3 vCPU wasted
Memory: 2GB →    16 GB      →    ~14 GB wasted
Disk: 20 GB →    500 GB     →    ~480 GB wasted

Step 3: Right-Size

Compute

• Monitor actual CPU and memory usage for 2 weeks
• Choose instance type that matches the 95th percentile usage + 20% buffer
• Consider burstable instances (t3/​t4g) for variable workloads

Database

• Use read replicas for read-heavy workloads instead of scaling up
• Consider serverless options (Aurora Serverless, PlanetScale) for variable traffic
• Downsize dev/​staging databases — they don't need production specs

Storage

• Use appropriate storage tiers:
  • Hot: Frequently accessed (SSD, S3 Standard)
  • Warm: Occasional access (S3 IA, cheaper disk)
  • Cold: Rarely accessed (S3 Glacier, archive)
  • Delete: Never accessed (set lifecycle policies)

Pricing Strategies

Reserved / Committed Use

When to Use Spot Instances

• CI/​CD runners
• Batch processing jobs
• Data pipeline workers
• Stateless web servers (with proper auto-scaling)
• Development/​testing environments

Never use Spot for: Databases, single-instance services, stateful workloads

Serverless Where It Fits

Consider serverless (Lambda, Cloud Functions, Cloud Run) when:

• Traffic is bursty (scale to zero when idle)
• Execution time < 15 minutes
• Requests are independent (no shared state)
• Cost per request < cost of always-on instances

Environment Optimization

Development / Staging

Auto-Scaling Scripts

# Stop dev environment at 7 PM
0 19 * * 1-5 kubectl scale deployment --all --replicas=0 -n dev

# Start dev environment at 8 AM
0 8 * * 1-5 kubectl scale deployment --all --replicas=1 -n dev

# AWS: Stop dev instances on weekends
aws ec2 stop-instances --instance-ids $(aws ec2 describe-instances \
  --filters "Name=tag:Environment,Values=dev" \
  --query "Reservations[].Instances[].InstanceId" --output text)

Network Cost Reduction

Data Transfer Costs (Often Overlooked)

• Same region — usually free between services
• Cross-region — $0.01-0.02/​GB — avoid unless necessary
• Internet egress — $0.09-0.12/​GB — use CDN for static assets
• NAT Gateway — $0.045/​GB — expensive! Use VPC endpoints for AWS services

Optimization

• Use a CDN (CloudFront, Cloudflare) for static content
• Compress API responses (gzip/​brotli)
• Use VPC endpoints for S3, DynamoDB, etc.
• Keep services in the same region/​zone when possible
• Use private networking instead of public IPs for internal communication

Cost Report Template

# Infrastructure Cost Report

**Period**: [Month/​Quarter]
**Total Spend**: $X,XXX
**Change from Last Period**: +/​-X%

## Top 5 Cost Drivers
| Service | Cost | % of Total | Trend |
|---------|------|-----------|-------|
| EC2/​Compute | $X | X% | ↑/↓/→ |
| RDS/​Database | $X | X% | ↑/↓/→ |
| S3/​Storage | $X | X% | ↑/↓/→ |
| Data Transfer | $X | X% | ↑/↓/→ |
| Other | $X | X% | ↑/↓/→ |

## Waste Identified
| Resource | Monthly Cost | Action |
|----------|-------------|--------|
| [specific resource] | $X | [downsize/​delete/​reserve] |

## Recommendations
1. [Action] — estimated savings: $X/​month
2. [Action] — estimated savings: $X/​month
3. [Action] — estimated savings: $X/​month

## Total Potential Savings: $X/​month (X% of current spend)

Key Metrics to Track

• Cost per request — are you getting cheaper or more expensive per unit?
• Cost per customer — infrastructure cost divided by active users
• Utilization rate — what % of provisioned capacity is actually used?
• Waste ratio — idle or over-provisioned resources as % of total spend
• Reserved coverage — what % of steady-state compute is reserved?