Step 3 — Storage & Databases

Storage and database questions account for a significant portion of the SAA-C03 exam. The key is knowing which service fits which scenario — not just what each service does, but when the exam expects you to choose it over the alternatives.

S3 — Simple Storage Service

S3 is object storage: you store files (objects) inside containers (buckets). No file system, no block device — just HTTP PUT/GET to a key-value store that scales to exabytes.

S3 Storage Classes (Hierarchy by Cost)

S3 Standard              — Frequent access. 11 9s durability. 3 AZ replication.
S3 Intelligent-Tiering   — Auto-moves objects between tiers based on access.
S3 Standard-IA           — Infrequent Access. Lower storage cost, retrieval fee.
S3 One Zone-IA           — Like Standard-IA but only 1 AZ. 20% cheaper, less resilient.
S3 Glacier Instant       — Archive, millisecond retrieval.
S3 Glacier Flexible      — Archive, minutes-to-hours retrieval.
S3 Glacier Deep Archive  — Cheapest. 12-hour retrieval. Compliance/long-term backup.

S3 Intelligent-Tiering automatically moves objects that haven’t been accessed for 30+ days into cheaper tiers — it’s the low-management answer for mixed-access workloads.

S3 Lifecycle Policies

Automate transitions and expiration:

Upload ──► S3 Standard (0 days)
              │
              ▼ after 30 days
          S3 Standard-IA
              │
              ▼ after 90 days
          S3 Glacier Instant Retrieval
              │
              ▼ after 365 days
          S3 Glacier Deep Archive
              │
              ▼ after 2555 days (7 years)
          Delete (expired)

S3 Key Features

Versioning — Keeps all versions of every object. Once enabled on a bucket, you can only suspend (not disable) it. Protects against accidental deletes and overwrites.

Replication — Cross-Region Replication (CRR) copies objects to a bucket in a different region (compliance, lower latency for remote users). Same-Region Replication (SRR) copies within the same region (log aggregation, live replication between accounts). Requires versioning on both buckets.

S3 Transfer Acceleration — Routes uploads through CloudFront edge locations → AWS backbone → destination bucket. Speeds up large uploads from globally distributed users.

Multipart Upload — Required for objects > 5 GB, recommended for > 100 MB. Enables parallel upload of parts.

Static Website Hosting — S3 can host a static website (HTML/CSS/JS) directly. No EC2 needed. Combine with CloudFront for HTTPS and global edge caching.

S3 Security

Bucket Policy         — Resource-based JSON policy. Controls who can access the bucket.
ACLs                  — Legacy. Avoid unless required for specific cross-account cases.
Block Public Access   — Account-level or bucket-level override. Prevents accidental exposure.
Server-Side Encryption:
  SSE-S3   — AWS manages keys (AES-256). Default since Jan 2023.
  SSE-KMS  — You manage keys in KMS. Audit trail. Extra cost.
  SSE-C    — You provide keys per request. AWS never stores your key.

Presigned URLs — Temporarily grant access to a private object. Time-limited (default 1 hour, max 7 days with STS). The exam loves this for “allow a third party to download a private object without making it public.”

Block & File Storage: EBS, Instance Store, EFS

EBS — Elastic Block Store

Network-attached block storage for EC2. Persists independently of instance lifecycle (unlike instance store).

gp3 (General Purpose SSD)   — Default. 3,000 IOPS baseline, up to 16,000 IOPS.
io2 Block Express            — Highest performance. Up to 256,000 IOPS. Multi-Attach.
st1 (Throughput HDD)         — Sequential reads. Big data, log processing. Low cost.
sc1 (Cold HDD)               — Cheapest. Infrequent access archival.

EBS Multi-Attach — io2 volumes can attach to multiple EC2 instances in the same AZ simultaneously. Requires cluster-aware filesystem.

EBS Snapshots — Point-in-time backup stored in S3 (but you access via EBS APIs). Incremental after first snapshot. Can copy snapshots cross-region for DR.

Instance Store

Physically attached NVMe SSDs on the EC2 host. Fastest possible storage (sub-millisecond), but ephemeral — data is lost when the instance stops, terminates, or the underlying host fails. Use for temporary data: cache, buffers, scratch space.

EFS — Elastic File System

Managed NFS file system mountable by multiple EC2 instances simultaneously — even across AZs. This is the key differentiator from EBS (which is single-AZ by default).

EBS                           EFS
─────────────────────         ────────────────────────────
Single EC2 instance           Multiple instances + AZs
Single AZ                     Regional (multi-AZ)
Block storage                 File system (NFS)
Fixed provisioned size        Scales automatically
Faster for single-instance    Better for shared workloads

EFS Infrequent Access (EFS-IA) — Automatically moves files not accessed for 30+ days to a cheaper storage tier. Savings up to 92% vs standard.

RDS — Relational Database Service

RDS manages the database engine for you: patching, backups, storage scaling, and failover. You choose the engine, AWS handles the rest.

Supported Engines

MySQL, PostgreSQL, MariaDB, Oracle, Microsoft SQL Server, and Amazon Aurora.

Multi-AZ vs Read Replicas (Most Tested RDS Topic)

Multi-AZ                              Read Replicas
──────────────────────────────────    ─────────────────────────────────────
Synchronous replication               Asynchronous replication
Same region (different AZ)            Same region, cross-region, or cross-account
Standby is NOT accessible             Read replica IS accessible (read traffic)
Automatic failover (~60 seconds)      No automatic failover
Purpose: HIGH AVAILABILITY            Purpose: READ SCALABILITY + REPORTING
DNS endpoint stays the same           Different endpoint per replica

Exam rule: If the question mentions “failover”, “disaster recovery”, “availability” → Multi-AZ. If it mentions “read-heavy”, “analytics offload”, “reporting” → Read Replica.

Multi-AZ can be in a different AZ but not a different region by default. For cross-region HA, use Aurora Global Database or RDS cross-region read replicas (then promote in a DR event).

Amazon Aurora

AWS’s proprietary database engine, compatible with MySQL and PostgreSQL but architecturally different:

Aurora Storage: distributed across 3 AZs, 6 copies of data, auto-heals
Aurora Compute: up to 15 read replicas (vs 5 for regular RDS)
Aurora Failover: ~30 seconds (vs ~60 for regular RDS Multi-AZ)
Aurora Cost: ~20% more than RDS, but 5x better throughput

Aurora Serverless v2 — Scales compute in fine-grained increments based on actual load. Ideal for variable or unpredictable workloads, dev/test environments. Billed per ACU-second (Aurora Capacity Unit).

Aurora Global Database — One primary region, up to 5 read-only secondary regions. Replication lag < 1 second. In a regional disaster, promote a secondary to primary in < 1 minute. This is the exam answer for “cross-region DR with sub-second RPO for relational data.”

DynamoDB — Serverless NoSQL

DynamoDB is a fully managed key-value and document database. No servers to manage, no schema migrations, single-digit millisecond performance at any scale.

Data Model

Every item has a partition key (required). Optionally add a sort key (creates a composite primary key enabling range queries).

Table: Orders
Partition Key: CustomerId
Sort Key: OrderDate

CustomerId  │ OrderDate   │ Total  │ Status
────────────┼─────────────┼────────┼────────
user_123    │ 2026-01-15  │ $142   │ Shipped
user_123    │ 2026-03-01  │ $67    │ Pending
user_456    │ 2026-02-20  │ $890   │ Delivered

Capacity Modes

Provisioned — You set Read Capacity Units (RCU) and Write Capacity Units (WCU). Enable auto-scaling to adjust automatically. Best for predictable, sustained traffic.

On-Demand — Pay per request. No capacity planning needed. Best for variable, spiky, or unknown traffic patterns. More expensive per request but no waste.

Secondary Indexes

Global Secondary Index (GSI) — Query on any non-key attribute. Has its own partition/sort key. Queries eventually consistent. Can be added/removed after table creation.

Local Secondary Index (LSI) — Same partition key, different sort key. Strongly consistent queries possible. Must be defined at table creation. Max 10 GB per partition key value.

DynamoDB Accelerator (DAX)

In-memory cache specifically for DynamoDB. Reduces read latency from milliseconds to microseconds. API-compatible — no application code changes needed. Use for read-heavy, latency-sensitive workloads.

DynamoDB Streams

Captures a time-ordered sequence of item modifications. Enables event-driven architectures:

DynamoDB Table ──► DynamoDB Stream ──► Lambda Function ──► SNS / SQS / ElasticSearch
                   (24-hour window)

Common exam use: “Trigger a process whenever an item is added or modified in a DynamoDB table” → DynamoDB Streams + Lambda.

Storage Decision Matrix for the Exam