Nerd Level Tech logo
|
Lesson 18 of 24

Security, Cost & Well-Architected Frameworks

Cost Optimization Strategies

4 min read

Cost optimization is a key differentiator for cloud architects. Interviewers expect you to balance performance, reliability, and cost effectively.

AWS Pricing Models

Compute Pricing Options

Model Discount Commitment Best For
On-Demand 0% None Variable workloads
Savings Plans Up to 72% 1-3 years Predictable compute
Reserved Instances Up to 72% 1-3 years Specific instances
Spot Instances Up to 90% None Fault-tolerant workloads

Savings Plans vs Reserved Instances

Aspect Savings Plans Reserved Instances
Flexibility Any instance family Specific instance type
Region Region-flexible (Compute SP) Region-specific
Service EC2, Lambda, Fargate EC2, RDS only
Management Simpler More complex
Recommendation Modern choice Legacy workloads

Interview Question: Cost Strategy

Q: "Design a cost optimization strategy for a variable workload with a predictable baseline."

A: Hybrid approach:

Workload Analysis:
├── Baseline: 100 instances (consistent)
├── Peak: 300 instances (variable)
└── Fault-tolerant batch: 50 instances

Cost Strategy:
├── Baseline (100): Compute Savings Plan (3-year)
│   └── 66% savings
├── Variable (0-200): On-Demand + Spot mix
│   ├── Critical: On-Demand
│   └── Non-critical: Spot (70/30 mix)
└── Batch (50): Spot Instances
    └── Up to 90% savings

Expected Savings: 45-55% compared to all On-Demand.

Right-Sizing

Analysis Process

  1. Collect Metrics (2-4 weeks minimum)

    • CPU utilization
    • Memory usage
    • Network throughput
    • Storage IOPS

  2. Identify Candidates

    • Average CPU < 40%: Downsize
    • CPU consistently > 80%: Upsize
    • Memory-bound: Consider memory-optimized

  3. Test and Validate

    • Load test new size
    • Monitor for 1-2 weeks
    • Rollback plan ready

AWS Tools for Right-Sizing

Tool Function Cost
Cost Explorer Right-sizing recommendations Free
Compute Optimizer ML-based recommendations Free (basic)
Trusted Advisor Instance recommendations Business/Enterprise

Interview Question: Right-Sizing RDS

Q: "Your RDS database is consistently at 20% CPU. What's your recommendation?"

A: Don't immediately downsize. Analyze holistically:

  1. Check Memory Pressure

    • Low CPU doesn't mean oversized
    • Database may be memory-bound

  2. Analyze Query Patterns

    • Peak vs average utilization
    • Batch job timing

  3. Consider Buffer Pool

    • Larger instance = more cache
    • May improve performance significantly

  4. Recommendation

    • If memory and IOPS also low: Downsize
    • If memory high: Stay or switch to memory-optimized
    • Test thoroughly before production change

Storage Optimization

S3 Storage Classes

Class Access Pattern Cost (per GB) Retrieval
Standard Frequent $$$ Instant
Intelligent-Tiering Variable $$$ + monitoring Instant
Standard-IA Infrequent (30+ days) $$ Instant
One Zone-IA Infrequent, non-critical $ Instant
Glacier Instant Archive, instant access $ Instant
Glacier Flexible Archive ¢ Minutes-hours
Glacier Deep Archive Long-term archive ¢ Hours

S3 Lifecycle Policies

Lifecycle Rules:
  - Transition to Standard-IA: 30 days
  - Transition to Glacier: 90 days
  - Transition to Deep Archive: 365 days
  - Delete: 7 years (compliance)

EBS Optimization

Volume Type Use Case Cost Consideration
gp3 General purpose 20% cheaper than gp2
io2 High IOPS Pay for provisioned IOPS
st1 Throughput Cheaper for sequential
sc1 Cold storage Cheapest block storage

Quick Win: Migrate gp2 to gp3 for immediate 20% savings with same or better performance.

Data Transfer Costs

Transfer Cost Matrix

From To Cost
Internet AWS Free
AWS Internet $0.09/GB (first 10TB)
Same Region (AZ to AZ) $0.01/GB
Cross-Region $0.02/GB
VPC Peering (same region) $0.01/GB
PrivateLink $0.01/GB + hourly

Cost Reduction Strategies

  1. Use VPC Endpoints

    • S3 Gateway endpoint: Free
    • Avoid NAT Gateway data processing

  2. Regional Deployment

    • Keep compute near data
    • Multi-region only when required

  3. CloudFront for Egress

    • Often cheaper than direct egress
    • Additional caching benefits

  4. Compression

    • Compress data before transfer
    • Significant savings for text/logs

Interview Question: Data Transfer

Q: "Your monthly data egress is $50,000. How would you reduce it?"

A: Multi-pronged approach:

Strategy Potential Savings
CloudFront distribution 20-40%
Response compression 30-50%
Caching headers 20-30%
Regional endpoints 10-20%
S3 Transfer Acceleration review Variable

Action Plan:

  1. Analyze CloudWatch for transfer patterns
  2. Implement CloudFront for repeat requests
  3. Enable gzip/brotli compression
  4. Review API response sizes
  5. Consider reserved capacity for predictable egress

FinOps Practices

Cost Allocation

  1. Tagging Strategy

    • Mandatory: Environment, Owner, Project, CostCenter
    • Enforce via SCP or AWS Config

  2. AWS Organizations

    • Separate accounts by business unit
    • Consolidated billing
    • Reserved capacity sharing

Monitoring and Governance

Tool Purpose
AWS Budgets Alerts and forecasting
Cost Anomaly Detection ML-based anomaly alerts
Cost Explorer Analysis and reporting
Savings Plans recommendations Purchase guidance

Chargeback Model

Central IT Budget
    ↓
Cost Allocation Tags
    ↓
Monthly Reports per Team
    ↓
Business Unit Chargeback

Key Insight: Cost optimization is continuous, not one-time. Implement automated policies, regular reviews, and team accountability to maintain efficiency as workloads evolve.

Next, we'll explore the AWS Well-Architected Framework. :::

Quiz

Module 5: Security, Cost & Well-Architected Frameworks

Take Quiz