AWS SAA Cheat Sheet
Quick-reference summary of the key facts for the SAA-C03 exam.
Exam at a glance
- Exam code: SAA-C03 · Fee: $150 USD · Questions: 65 (50 scored + 15 unscored) · Time: 130 minutes · Pass: 720 scaled (100–1,000) · Valid: 3 years · Retake: 14-day wait, full fee
- Domains: Secure 30% · Resilient 26% · High-Performing 24% · Cost-Optimized 20%
Domain 1: Design Secure Architectures (30%)
- IAM: Roles (temporary credentials) over users with access keys, always. Least privilege. Explicit deny beats any allow. Root user: MFA on, no access keys, never for daily work. IAM Identity Center for workforce SSO.
- Security groups vs NACLs: SG = stateful, instance-level, allow-only. NACL = stateless, subnet-level, numbered allow/deny rules. Both must allow traffic.
- VPC: Public subnets (IGW route) for load balancers only; private subnets for app + DB. NAT gateway per AZ for outbound-only internet from private subnets. Gateway endpoint = S3/DynamoDB (free, route table). Interface endpoint = other services (PrivateLink).
- Encryption: SSE-KMS with customer managed key when key control + audit trail required. Secrets Manager rotates secrets automatically; Parameter Store does not.
- WAF vs Shield: WAF filters application-layer requests (SQLi, XSS). Shield defends against DDoS. GuardDuty detects threats. CloudTrail audits API calls (who did what). CloudWatch monitors metrics/logs/alarms. Config checks resource configuration against rules.
Domain 2: Design Resilient Architectures (26%)
- Multi-AZ = high availability (automatic failover). Read replicas = read scaling. Never confuse them.
- Decoupling: SQS = queue, one consumer per message, buffers spikes. SNS = pub/sub fan-out, every subscriber gets the message. Dead-letter queue catches poison messages. Visibility timeout must exceed processing time.
- Route 53: Alias record at zone apex (CNAME cannot go there). Weighted = traffic splits. Latency-based = lowest-latency Region. Geolocation = by user location. Failover = primary/standby with health checks. DNS failover is not instant (TTL caching).
- DR: RTO = how long you can be down. RPO = how much data you can lose. Backup & restore (hours, cheapest) → pilot light (tens of minutes) → warm standby (minutes) → active-active (near zero, priciest).
- S3 durability: 99.999999999% (eleven 9s) across AZs. Versioning is the prerequisite for replication and the fix for accidental overwrite/delete.
Domain 3: Design High-Performing Architectures (24%)
- EC2 families: C = compute, R = memory, I/D = storage, T = burstable (CPU credits; bad for sustained load). Cluster placement group = lowest latency, single AZ. Spread = hardware isolation (7/AZ). Partition = rack isolation for large clusters.
- Lambda: 15-minute max execution. Memory 128 MB–10 GB (CPU scales with memory). Provisioned concurrency kills cold starts. Needs NAT gateway for internet when VPC-attached.
- Databases: RDS Multi-AZ = failover; read replicas (5 max) = reads. Aurora = up to 15 replicas, 6-way storage replication across 3 AZs, seconds failover, auto-scaling storage to 128 TB. DynamoDB = key-value, single-digit ms; on-demand for spiky, provisioned for predictable; DAX = microsecond cache; global tables = multi-Region multi-active. ElastiCache absorbs repeated reads.
- CloudFront vs Global Accelerator: CloudFront caches content at the edge. Global Accelerator improves the network path for non-cacheable dynamic traffic + fast Region failover with static IPs.
- Load balancers: ALB = Layer 7, path/host routing. NLB = Layer 4, static IPs, extreme throughput. GWLB = third-party virtual appliances.
Domain 4: Design Cost-Optimized Architectures (20%)
- Compute: On-Demand = flexible, priciest. Reserved/Savings Plans = steady baseline, up to ~72% off. Spot = interruptible batch, up to ~90% off. Compute Savings Plan = flexible across families/Regions.
- Storage: Lifecycle policies move aging data down the S3 ladder and expire it. Intelligent-Tiering when access is unpredictable. gp3 baseline 3,000 IOPS; io2 for sustained high IOPS.
- Data transfer: Inbound free; outbound charged. CloudFront + VPC endpoints cut egress. Cross-AZ traffic is charged — keep chatty tiers in the same AZ where possible.
- Cost tools: Cost Explorer = analyze spend. Budgets = alerts on forecast/actual. Compute Optimizer = right-sizing. Trusted Advisor = waste + best practices.
S3 storage classes
| Class | Use when | Availability | Min. duration |
|---|---|---|---|
| Standard | Frequent access | 99.99% | None |
| Intelligent-Tiering | Access pattern unknown/changing | 99.9% | None |
| Standard-IA | Infrequent, needs instant access | 99.9% | 30 days |
| One Zone-IA | Infrequent, re-creatable data | 99.5% | 30 days |
| Glacier Instant Retrieval | Archive, quarterly access, ms retrieval | 99.9% | 90 days |
| Glacier Flexible Retrieval | Archive, 1–2 accesses/year | 99.99% | 90 days |
| Glacier Deep Archive | Long-term archive, rarely accessed | 99.99% | 180 days |
Glacier retrieval: Flexible — Expedited 1–5 min, Standard 3–5 hrs, Bulk 5–12 hrs (free). Deep Archive — Standard within 12 hrs, Bulk within 48 hrs; no Expedited.
EBS vs EFS vs instance store
| Service | Scope | Shared? | Survives stop? |
|---|---|---|---|
| EBS | Block, one AZ | One instance (Multi-Attach: same-AZ instances, io1/io2) | Yes |
| EFS | File (NFS), multi-AZ | Yes, many instances | Yes |
| Instance store | Block, on host | No | No — lost on stop/terminate/host failure |
| FSx for Windows | File (SMB), AD-integrated | Yes | Yes |
RDS vs Aurora vs DynamoDB
| Service | Model | Best for | Scaling |
|---|---|---|---|
| RDS | Relational (MySQL, PostgreSQL, Oracle, SQL Server, MariaDB) | Traditional relational workloads, any engine | Read replicas (5), vertical scaling |
| Aurora | Relational, MySQL/PostgreSQL-compatible | High availability, many replicas, fast failover | 15 read replicas, auto storage to 128 TB |
| DynamoDB | NoSQL key-value/document | Massive scale, single-digit-ms key lookups | Automatic; on-demand or provisioned |
Must-know numbers
- S3 durability 99.999999999% (eleven 9s). Max object size 5 TB.
- Lambda max execution 15 minutes; max memory 10 GB; /tmp 512 MB default (up to 10 GB).
- EC2 Spot discount up to 90%; Reserved/Savings Plans up to ~72%.
- gp3 baseline 3,000 IOPS, 125 MiB/s throughput. io2 up to 64,000 IOPS per volume (Nitro instances).
- SQS standard: nearly unlimited throughput, at-least-once delivery. FIFO: 300 messages/sec (3,000 with batching), exactly-once processing, strict ordering.
- RDS automated backup retention up to 35 days. DynamoDB point-in-time recovery: 35 days.
- CloudFront invalidations: first 1,000 paths/month free.
- SAA-C03: 65 questions, 130 minutes = exactly 2 minutes per question.
Advertisement