</>Học Dev
Bài học

Tuần 11 - Ngày 2: Review Domain 3 & 4

Tuần 11 – Ngày 2

Tuần 11 - Ngày 2: Review Domain 3 & 4

Mục tiêu

Tổng ôn 2 domain còn lại (chiếm 44% của exam):

  • Domain 3: Design High-Performing Architectures (24%)
  • Domain 4: Design Cost-Optimized Architectures (20%)

Domain 3: Design High-Performing Architectures (24%)

3.1 Storage performance

Block storage (EBS)

  • gp3: general purpose, 3000 IOPS baseline (provision up to 16,000)
  • io2 Block Express: highest perf (256,000 IOPS, 4 GB/s)
  • st1: HDD throughput (big data, logs)
  • sc1: cold HDD

File storage

  • EFS: NFS Linux, multi-AZ, auto-scale
  • FSx for Lustre: HPC, hundreds GB/s throughput
  • FSx for Windows: SMB, AD integration
  • FSx for ONTAP: multi-protocol enterprise
  • FSx for OpenZFS: NFS ZFS

Object storage (S3)

  • Standard: hot data
  • Intelligent-Tiering: unknown patterns
  • Express One Zone: < 10ms latency
  • Glacier tiers: archive
  • Transfer Acceleration: faster global upload
  • Multi-Region Access Points: global active-active

3.2 Compute performance

EC2 sizing

  • T-family: burstable (general web)
  • M-family: balanced
  • C-family: compute-optimized
  • R-family: memory-optimized
  • G/P-family: GPU
  • I-family: storage (NVMe SSD)

Serverless

  • Lambda: scale per request, 15-min limit
  • Fargate: containers, no server mgmt
  • Auto Scaling: target tracking, predictive

3.3 Database performance

WorkloadService
Relational OLTPRDS, Aurora
Relational + read-heavyAurora với readers / RDS read replicas
NoSQL key-valueDynamoDB
CachingElastiCache (Redis), DAX
Analytics (OLAP)Redshift
SearchOpenSearch
Time-seriesTimestream
LedgerQLDB

3.4 Network performance

Optimization tools

  • CloudFront: edge caching, global reach
  • Global Accelerator: anycast static IPs, AWS backbone
  • VPC Endpoints: avoid internet, lower latency
  • Direct Connect: dedicated low-latency hybrid
  • Transit Gateway: hub for multi-VPC

Placement

  • Cluster placement group: low-latency between instances
  • Edge locations: cache near users

Domain 4: Design Cost-Optimized Architectures (20%)

4.1 Cost-effective storage

S3 lifecycle

Standard → Standard-IA (30 days) → Glacier Instant (90 days) → Deep Archive (365 days) → Delete (7 years)

EBS optimization

  • gp2 → gp3 (~20% saving)
  • Delete unattached volumes
  • Snapshot lifecycle (DLM)
  • Snapshot Archive (90+ days)

Data transfer optimization

  • VPC Endpoints (no NAT)
  • CloudFront (cheaper egress)
  • Same-region traffic (free private IP)
  • DataSync (efficient transfer)

4.2 Cost-effective compute

Pricing models

  • On-Demand: 0% off (flexible)
  • Reserved Instance: up to 72% off (1-3y commit)
  • Savings Plans: up to 72% off (flexible commit)
  • Spot: up to 90% off (fault-tolerant)
  • Dedicated Host: BYOL, compliance

Right-sizing

  • Compute Optimizer recommendations
  • ARM Graviton (~20% cheaper)
  • Lambda Power Tuning

4.3 Cost-effective databases

  • Aurora Serverless v2: pay per ACU
  • DynamoDB On-Demand: no commitment
  • Reserved DynamoDB capacity: for known throughput
  • Right-size RDS (use Performance Insights)
  • Lifecycle to Glacier for old data

4.4 Cost monitoring

  • Cost Explorer: visualize + forecast
  • Budgets: alerts + actions
  • Cost Anomaly Detection: ML detect spikes
  • CUR: detailed analysis with Athena
  • Tags + Cost Allocation: track per project/team

Common Performance Patterns

Pattern 1: Web app scaling

  • ALB + ASG + Aurora
  • CloudFront for static + caching
  • ElastiCache for session/data
  • Auto-scaling based on metrics

Pattern 2: Big data

  • S3 data lake
  • EMR / Glue for processing
  • Redshift for warehouse
  • Athena for ad-hoc

Pattern 3: ML

  • S3 for training data
  • SageMaker for training (Spot for cost)
  • Lambda for inference
  • DynamoDB / OpenSearch for feature store

Pattern 4: Global app

  • CloudFront / Global Accelerator
  • DynamoDB Global Tables / Aurora Global
  • S3 CRR / MRAP
  • Route 53 latency routing

Common Cost Optimization Patterns

Strategy 1: Tier storage

  • Hot data: Standard
  • Warm: Standard-IA
  • Cold: Glacier
  • Archive: Deep Archive
  • Auto via lifecycle rules

Strategy 2: Compute mix

  • Baseline 24/7: Reserved 3y
  • Variable: Savings Plans
  • Burst peak: On-Demand
  • Batch: Spot

Strategy 3: Right-size + cleanup

  • Compute Optimizer EC2 recommendations
  • Delete idle resources (Trusted Advisor)
  • Unused EBS, snapshots
  • Idle Load Balancers, EIPs

Strategy 4: Cache + CDN

  • CloudFront caches → reduce origin
  • ElastiCache → reduce DB queries
  • DAX → reduce DynamoDB reads

Strategy 5: Serverless when low traffic

  • Lambda free tier 1M req/month
  • DynamoDB On-Demand
  • Pay-per-use cheaper than always-on

Quick Lookup Tables

Storage costs (us-east-1)

Service$/GB-month
S3 Standard$0.023
S3 Standard-IA$0.0125
S3 Glacier Instant$0.004
S3 Glacier Deep$0.00099
EBS gp3$0.08
EBS io2$0.125
EFS Standard$0.30
EFS One Zone-IA$0.0133
FSx Windows$0.13
FSx Lustre$0.140

Compute pricing tiers

StrategyDiscountTerm
On-Demand0%
Compute SP 1-year~30-50%1y
EC2 RI 1y NURI~30%1y
EC2 RI 1y AURI~36%1y
Compute SP 3-year~55-66%3y
EC2 RI 3y AURI~63-72%3y
SpotUp to 90%None

Performance Decision Trees

Right database?

Relational?OLTP,simpleRDSOLTP,high-perfAuroraAnalytics(OLAP)RedshiftNoSQL?Key-value,scalableDynamoDBDocumentDynamoDBorDocumentDBGraphNeptuneCacheElastiCacheSpecialty?Time-seriesTimestreamLedgerQLDBSearchOpenSearch

Right compute?

Event-driven,<15min?LambdaContainers?ServerlessFargateEC2-managedECS/EKSonEC2Predictablebatch?SpotInstances(withcheckpointing)Long-runningstateful?EC2(ReservedorSavingsPlans)GPU/specialized?SpecificEC2instancetype

Common Exam Tricks

Multi-answer questions

  • "Choose 2" or "Choose 3"
  • Read all options carefully
  • Each option must be true on its own

Best vs Acceptable

  • Sometimes multiple options work
  • Choose most cost-effective OR least operational overhead OR most secure

Modern vs Legacy

  • gp3 > gp2 (newer, cheaper)
  • HTTP API > REST API for simple (cheaper)
  • Compute Savings Plans > EC2 Convertible RI (more flexible)
  • OAC > OAI (newer)
  • Application Migration Service > Server Migration Service (newer)

Final Tips

Before exam

  • Review Well-Architected pillars
  • Memorize service limits (RDS replicas: 5, Aurora: 15, etc.)
  • Practice scenario questions (timed)
  • Get 7+ hours sleep

During exam

  • Read carefully (2 min/question average)
  • Flag uncertain → return at end
  • Eliminate wrong answers first
  • Trust your prep

Mindset

  • 720/1000 to pass (~72%)
  • 50 scored + 15 unscored questions
  • Don't stress single questions
  • Aim 80%+ on practice → ready for real

Tiếp theo: Practice Exam 1 (65 câu)