FlashGenius Logo FlashGenius
SAA-C03 · Domain 4 · 20% of Exam

Design Cost-Optimized Architectures

EC2 Pricing · S3 Lifecycle · Spot Instances · Savings Plans · Cost Explorer · VPC Endpoints

Study with Practice Tests →

Design Cost-Optimized Architectures — Overview

Domain 4 is 20% of the SAA-C03 exam. Questions present a scenario and ask for the lowest-cost solution that still meets requirements. Master EC2 pricing models, S3 storage classes, serverless trade-offs, and AWS cost governance tools.

Domain 4 · 20% of Exam · ~13 Questions

SAA-C03 Domain Breakdown

#DomainWeight
1Design Secure Architectures30%
2Design Resilient Architectures26%
3Design High-Performing Architectures24%
4Design Cost-Optimized Architectures20%

Task Statements

TaskStatementKey Topics
4.1Cost-optimized storage solutionsS3 storage classes, lifecycle policies, EBS, EFS tiers
4.2Cost-optimized compute solutionsEC2 pricing models, Spot, Lambda, Fargate, right-sizing
4.3Cost-optimized database solutionsAurora Serverless, DynamoDB on-demand, ElastiCache, RDS RI
4.4Cost-optimized network architecturesNAT Gateway, VPC endpoints, Direct Connect, data transfer
Exam Strategy: Domain 4 questions almost always present a scenario with two requirements: meets functional need AND minimizes cost. The wrong answers often work technically but are expensive. Know when Spot is risky, when serverless wins, and when VPC Gateway Endpoints eliminate unnecessary charges.

What You'll Master

EC2 Pricing Models

On-Demand vs Reserved (Standard/Convertible) vs Savings Plans vs Spot vs Dedicated Hosts. Know the right model for each workload pattern and the discount each provides.

S3 Cost Optimization

Storage class cost ladder from Standard to Deep Archive. Lifecycle policies to automate transitions. Minimum storage duration charges. Retrieval cost trade-offs.

Serverless vs Always-On

Lambda pay-per-use vs always-on EC2. Fargate task-second billing vs EC2 capacity. When serverless wins and when sustained EC2 is cheaper for high-throughput workloads.

Database Cost Strategies

Aurora Serverless v2 for intermittent databases. DynamoDB on-demand vs provisioned break-even. ElastiCache to reduce DB instance size. Redshift pause/resume for dev clusters.

Network Cost Reduction

NAT Gateway data processing charges. VPC Gateway Endpoints (free) for S3/DynamoDB. Cross-AZ vs cross-Region transfer pricing. Direct Connect break-even vs VPN.

Cost Management Tools

Cost Explorer for visualization, Budgets for alerts, CUR for raw billing data, Trusted Advisor for idle resource checks, Compute Optimizer for ML-based right-sizing.

Key Services Covered

Amazon EC2 (all pricing models)  ·  AWS Savings Plans  ·  Amazon S3 (storage classes + lifecycle)  ·  AWS Lambda  ·  AWS Fargate  ·  Amazon Aurora Serverless  ·  Amazon DynamoDB (on-demand)  ·  Amazon ElastiCache  ·  AWS Cost Explorer  ·  AWS Budgets  ·  AWS Trusted Advisor  ·  AWS Compute Optimizer  ·  VPC Endpoints  ·  AWS Direct Connect  ·  Amazon CloudFront  ·  AWS Auto Scaling

Core Concepts — Design Cost-Optimized Architectures

Eight deep-dive concept cards covering every examinable cost optimization topic in Domain 4.

1. EC2 Purchasing Options

OptionCommitmentDiscountBest For
On-DemandNoneUnpredictable, short-term, dev/test
Standard RI1 or 3 yearsUp to 72%Steady-state, locked to family/region/OS
Convertible RI1 or 3 yearsUp to 54%Steady-state, need flexibility to change family/OS
Compute Savings Plan1 or 3 yearsUp to 66%EC2 + Lambda + Fargate, any family/region/OS
EC2 Instance SP1 or 3 yearsUp to 72%Specific instance family in a region
Spot InstancesNoneUp to 90%Fault-tolerant batch, stateless, HPC
Dedicated HostsOptional RIVariesBYOL software (Oracle, Windows Server), compliance
Dedicated InstancesNoneHardware isolated from other accounts, no BYOL need
  • Spot rule: 2-minute interruption warning. Use for: batch jobs, CI/CD, stateless web workers, HPC. NEVER for: databases, session state, critical always-on services.
  • Selection heuristic: steady-state production → Reserved/Savings Plans · spiky/unknown → On-Demand · batch/fault-tolerant → Spot · BYOL compliance → Dedicated Hosts
  • Savings Plans vs RIs: Savings Plans commit to $/hr spend; automatically apply to eligible usage. RIs are locked to specific instance attributes. Savings Plans are the preferred modern mechanism.

2. S3 Cost Optimization

Storage ClassApprox. $/GB/moMin DurationRetrieval Fee
S3 Standard$0.023NoneNone
S3 Standard-IA$0.012530 daysPer-GB
S3 One Zone-IA$0.0130 daysPer-GB
S3 Intelligent-TieringVaries by tierNoneNone (monitoring fee)
Glacier Instant Retrieval$0.00490 daysPer-GB
Glacier Flexible Retrieval$0.003690 daysPer-GB (Expedited/Standard/Bulk)
Glacier Deep Archive$0.00099180 daysPer-GB (Standard/Bulk)
  • Lifecycle policies: automate transitions (e.g., Standard → Standard-IA after 30 days → Glacier after 90 days → expire after 365 days); avoid manual management
  • Minimum storage charges: deleting an object early still incurs the full minimum duration charge — factor this into class selection for short-lived objects
  • Requester Pays: bucket owner pays storage; requester pays data transfer and request costs — use for sharing large public datasets
  • Multipart Upload: required for objects >5GB, recommended >100MB; enables parallel uploads and retry of failed parts; reduces cost of failed large uploads

3. Compute Cost Optimization

  • Right-sizing: match instance type to actual workload; use AWS Compute Optimizer (ML-based recommendations from CloudWatch metrics); over-provisioned instances are the most common hidden cost
  • Auto Scaling: scale in during low demand to stop paying for idle capacity; combine with target tracking policy (e.g., CPU at 50%) for efficient utilization
  • Serverless cost model: Lambda = pay per 100ms invocation + request count; no idle cost; cheaper than always-on EC2 for intermittent workloads; for sustained high-throughput, EC2 may be cheaper per compute unit
  • Fargate vs EC2 launch type: Fargate = no idle cost (pay per task CPU/memory-second), simpler management, no cluster management; EC2 = cheaper for predictable sustained loads, more control over instance type
  • Spot Fleet / mixed fleet: On-Demand base capacity + Spot for additional; use capacity-optimized allocation strategy; diversify across instance types and AZs for availability
  • AWS Batch with Spot: batch jobs are ideal Spot candidates (retryable); managed compute environments with Spot can achieve up to 90% savings for batch processing

4. Database Cost Optimization

  • Aurora Serverless v2: pay per ACU-second; scales from minimum to maximum ACU; ideal for intermittent, unpredictable, or dev/test databases; vs Aurora Provisioned = always-on cluster hours regardless of utilization
  • RDS reserved instances: same 1- or 3-year commitment model as EC2 RIs; significant savings for steady-state RDS workloads; select instance class, engine, and region at purchase time
  • DynamoDB On-Demand vs Provisioned: On-Demand = pay per WRU/RRU; Provisioned = set WCU/RCU with auto-scaling (~70% cheaper for predictable load); On-Demand = cheaper for very spiky or very low traffic patterns
  • DynamoDB cost tips: enable TTL (auto-delete expired items — no WCU consumed); compress items before storing; use projections in GSIs (only project needed attributes to reduce GSI storage and read cost)
  • ElastiCache for read offload: cache frequent reads from RDS/Aurora; reduce required DB instance size; Memcached = simpler caching, cheaper per node; Redis = more features (pub/sub, sorted sets, persistence)
  • Redshift reserved nodes: reserve 1 or 3 years for significant savings; use Pause/Resume for dev/test clusters (pay only when running, storage always charged)

5. Network Cost Optimization

Traffic TypeCost
Data IN to AWS (from internet)Free
Data OUT to internetCharged (tiered, first 1GB/month free)
Same AZ (same Region)Free
Cross-AZ (same Region)$0.01/GB each way
Cross-RegionVaries by Region pair
NAT Gateway processing$0.045/hr + $0.045/GB processed
VPC Gateway Endpoint (S3/DynamoDB)Free
VPC Interface Endpoint~$0.01/hr per AZ + $0.01/GB
  • VPC Gateway Endpoints: free endpoints for S3 and DynamoDB from private subnets; eliminates NAT Gateway data processing charges; high-impact cost saver for S3-heavy workloads
  • CloudFront: caches content at edge; CloudFront-to-origin pricing is lower than EC2 direct internet egress; reduces requests hitting origin and associated compute cost
  • Direct Connect break-even: DX has monthly port charge + lower per-GB transfer cost; VPN = cheap setup + hourly connection cost; DX typically breaks even at ~1TB/month for large data egress from AWS
  • NAT Instance vs NAT Gateway: NAT Instance = EC2 pricing (cheaper), self-managed, no built-in HA, single point of failure; NAT Gateway = managed, HA within AZ, scales automatically, costs more per hour

6. AWS Cost Management Tools

  • AWS Cost Explorer: visualize spending over time; filter by service, region, account, tag; identify trends and anomalies; rightsizing recommendations; RI utilization/coverage reports; hourly granularity available (extra cost)
  • AWS Budgets: set custom cost/usage/RI/Savings Plan coverage budgets; alert via SNS/email when threshold reached (actual or forecasted); can trigger automated actions (apply SCP, stop EC2) at budget breach
  • AWS Cost and Usage Report (CUR): most detailed billing data; line-item records for every resource; delivered to S3; query with Athena; used for chargeback, showback, and detailed multi-account analysis
  • AWS Trusted Advisor: automated best practice checks; cost optimization checks: idle EC2, underutilized EBS volumes, unassociated Elastic IPs, RI purchase recommendations; full checks require Business or Enterprise support plan
  • AWS Compute Optimizer: ML-based recommendations for EC2, EBS, Lambda, ECS on Fargate, Auto Scaling groups; analyzes CloudWatch metrics; identifies over-provisioned resources; helps right-size workloads
  • Cost allocation tags: user-defined tags (e.g., Project: Analytics, Team: DataEng); activate in Billing console; visible in CUR and Cost Explorer for chargeback and cost attribution

7. Well-Architected Cost Optimization Pillar

  • 5 design principles: Implement cloud financial management · Adopt a consumption model · Measure overall efficiency · Stop spending money on undifferentiated heavy lifting · Analyze and attribute expenditure
  • Expenditure awareness: use tagging strategy; set budgets; use AWS Organizations for consolidated billing; regular cost reviews; identify top spenders by service/team/project
  • Cost-effective resources: right service for the job (managed vs self-managed); right pricing model (Reserved vs Spot vs On-Demand); right size (Compute Optimizer)
  • Manage demand and supply: Auto Scaling matches capacity to demand; SQS buffers demand spikes to avoid over-provisioning for peak; CloudFront reduces origin load
  • Optimize over time: decommission unused resources (Trusted Advisor idle checks); quarterly architecture reviews; adopt new lower-cost AWS services as they launch
  • TCO considerations: include compute, storage, network, licenses, operations, support; AWS TCO Calculator for on-prem vs cloud comparison; factor in operational savings from managed services

8. Architecting for Cost-Efficient Workloads

  • Serverless-first: Lambda + DynamoDB + API Gateway for CRUD APIs — extremely cheap at low-to-medium scale; scales to zero; no idle infrastructure costs
  • S3 as backbone: use S3 for data storage (cheapest durable store); build event-driven pipelines with S3 Events → Lambda → further processing; avoid unnecessary EC2 as intermediary
  • Container vs Function trade-off: Lambda = 15-min limit, per-invocation billing, cold starts; Fargate = no time limit, predictable performance, per-second billing; EC2 = cheapest per sustained compute unit
  • Caching strategy: CloudFront (edge) → ElastiCache (middle tier) → read replicas (database tier); each caching layer reduces load and cost on the tier below
  • Use managed services: trade operational overhead for cost predictability; factor DBA time into EC2 self-managed MySQL TCO vs RDS; Aurora Serverless vs provisioned for variable load
  • Data lifecycle automation: automate S3 transitions via lifecycle policies; DynamoDB TTL for ephemeral data; schedule EBS snapshot deletion; set CloudWatch log retention to avoid unbounded storage growth

Memory Hooks

Six high-impact mnemonics to lock in the trickiest Domain 4 concepts before exam day.

Spot Use Rule
"90% off, 2-min warning — batch only"
Spot = up to 90% off but can be interrupted with 2-minute warning. Only use for: batch jobs, CI/CD, stateless web workers, HPC. NEVER for: databases, session state, or critical always-on services. If the workload can restart — Spot it.
💡
Savings Plans vs RIs
"Savings Plans = flexible · RIs = specific"
Savings Plans commit to $/hr spend and flex across any instance family/region/OS. RIs lock to specific attributes for higher discount. Key differentiator: Compute Savings Plans cover EC2 + Lambda + Fargate in one commitment — RIs do NOT cover Lambda.
🌐
Free Network Traffic
"INTO AWS = free · Same AZ = free · S3 endpoint = free"
INTO AWS from internet = free. Within same AZ = free. Cross-AZ = $0.01/GB each way. VPC Gateway Endpoints for S3/DynamoDB = free — eliminating NAT Gateway charges is a common high-value exam scenario.
🔧
Cost Tools Trio
"Explorer = see · Budgets = alert · CUR = detail"
Cost Explorer = visualize and trend historical spending. Budgets = forward-looking alerts when actual or forecasted spend exceeds threshold (can trigger automated actions). CUR = raw line-item detail for Athena analysis and chargeback. Use all three for complete cost governance.
🏛️
NAT Gateway Savings
"High S3 traffic? Add a Gateway Endpoint (free!)"
NAT Gateway charges $0.045/hr + $0.045/GB processed — adds up fast with heavy S3 traffic from private subnets. Solution: create a Gateway VPC Endpoint for S3 (free) and update the route table. Removes NAT GW from the data path entirely. Classic exam scenario.
🗄️
DynamoDB Mode Selection
"Predictable = Provisioned · Spiky = On-Demand"
Predictable, consistent traffic → Provisioned (+ auto-scaling) = ~70% cheaper per request. Spiky, unknown, low-traffic, or dev/test → On-Demand = no capacity planning, no throttling risk. Also: enable TTL for free deletes — no WCU consumed when DynamoDB removes expired items.

Practice Quiz — Domain 4: Design Cost-Optimized Architectures

10 scenario-based questions. Select your answers, then click Submit Quiz to see your score.

Q1. A company runs a batch image processing workload that can tolerate interruptions and restarts. The job runs for 4 hours every night. What is the MOST cost-effective EC2 purchasing option?
Q2. A company needs persistent, steady-state EC2 capacity running 24/7 for a production database over the next 3 years. They want maximum cost savings. Which purchasing option provides the HIGHEST discount?
Q3. A company has EC2 instances that are over-provisioned but the team does not know by how much. Which AWS tool provides ML-based rightsizing recommendations based on actual CloudWatch utilization metrics?
Q4. A development team has EC2 instances for a staging environment that are only needed Monday–Friday, 8 AM–6 PM. What is the MOST cost-effective approach to eliminate weekend and evening costs?
Q5. A company runs EC2 instances in private subnets that frequently access Amazon S3. They notice high NAT Gateway data processing charges. What is the BEST way to eliminate these charges?
Q6. A startup wants to run an API backend that handles 10,000 requests per day with minimal operational overhead. The requests average 500ms each. What is likely the MOST cost-effective approach?
Q7. A company receives a large AWS bill and wants to identify which team or project is responsible for the highest costs. What should they implement?
Q8. A company uses Aurora MySQL for a dev/test database that is only accessed occasionally during business hours. What is the MOST cost-effective database option?
Q9. A company transfers 5TB of data per month from EC2 instances to an on-premises data center over the internet. The transfer costs are higher than expected. What should they evaluate?
Q10. An architect needs to set up automated alerts when the monthly AWS bill for a specific account is forecasted to exceed $10,000. Which service should they use?

Flashcards

Click any card to flip it and reveal the answer. Click again to flip back.

Click a card to flip it · Click again to flip back

Spot Instance Interruption
What happens when AWS needs the capacity back?
You receive a 2-minute interruption notice via instance metadata and EventBridge. Use lifecycle hooks or Spot interruption handlers to checkpoint state before termination. Only use Spot for fault-tolerant, restartable workloads.
Standard RI vs Convertible RI
Key trade-off between the two Reserved Instance types
Standard RI = up to 72% discount, locked to instance family/OS/region — maximum savings, minimum flexibility.

Convertible RI = up to 54% discount, can exchange for different family/OS/tenancy — more flexibility, less discount.
VPC Gateway vs Interface Endpoint
Which type is free, and what services does each support?
Gateway Endpoint (S3 + DynamoDB) = FREE — uses route table entry, no hourly charge, no data charge.

Interface Endpoint (PrivateLink, 150+ services) = ~$0.01/hr per AZ + $0.01/GB data charge.
AWS Compute Optimizer
What data does it analyze and what does it recommend?
Analyzes CloudWatch metrics (CPU, memory via CW agent, network, disk) over 14 days. Recommends optimal EC2 instance size, EBS volume type, Lambda memory allocation, and Auto Scaling group configurations to eliminate over-provisioning.
DynamoDB TTL and WCU Cost
Does deleting expired items via TTL consume WCU?
No — TTL deletions are free (no WCU consumed). Items are deleted within 48 hours of expiry. Use TTL for sessions, caches, and time-bound data to avoid manual deletion costs. Expired items stop being returned in queries immediately.
S3 Minimum Storage Duration
What happens if you delete an object before the minimum?
You are charged for the full minimum duration regardless. Minimums: Standard-IA / One Zone-IA = 30 days; Glacier Instant Retrieval = 90 days; Glacier Flexible = 90 days; Glacier Deep Archive = 180 days.
Savings Plans vs RIs — Lambda Coverage
Which commitment type covers AWS Lambda?
Compute Savings Plans cover Lambda (and Fargate and EC2). Reserved Instances do not cover Lambda. Savings Plans are more flexible and the preferred modern commitment mechanism for mixed compute workloads.
Cost Explorer vs AWS Budgets
What is the key functional difference?
Cost Explorer = historical analysis and visualization — what has already happened. Trends, filters, rightsizing reports.

Budgets = forward-looking alerts — notify when actual or forecasted spend is projected to exceed a threshold. Can trigger automated actions.

Study Advisor

Select your experience level to get a focused study plan for Domain 4.

Beginners

    Study Resources

    Official and recommended resources for mastering AWS cost optimization for the SAA-C03 exam.

    Ready to Ace the AWS SAA-C03 Cost Optimization Domain?

    Practice with full-length mock exams and adaptive flashcards on FlashGenius

    Start Free Practice →