Where orders are kept

Run it yourself, or let AWS run it?

You can install a database yourself on an EC2-hosted database: an Amazon Elastic Compute Cloud (Amazon EC2) instance where you patch the operating system (OS), install the engine, take backups and set up failover. Or you can use an AWS managed database, where AWS handles hardware, patching, backups and high availability, and you look after your data, users and settings.

Sam: Managed means no more 2 a.m. database backups by hand. I'm listening.

Exam tip Pick EC2-hosted only when you need OS-level access or an engine or version AWS doesn't offer. "Least operational overhead" on the exam = a managed database.

Relational: Amazon RDS and Amazon Aurora

A relational database stores data in tables of rows and columns that link to each other, and you query it with Structured Query Language (SQL). Orders, customers and stock fit this shape. Amazon Relational Database Service (Amazon RDS) runs familiar engines for you, such as MySQL, PostgreSQL, MariaDB, Oracle and Microsoft SQL Server. Amazon Aurora is AWS's own MySQL- and PostgreSQL-compatible engine, built for the cloud, keeping six copies of your data across three Availability Zones (AZs).

Exam tip Trap: Aurora is compatible with MySQL and PostgreSQL only. A managed SQL Server or Oracle database = Amazon RDS.

NoSQL: Amazon DynamoDB

NoSQL ("not only SQL") databases skip fixed tables for flexible shapes, such as key-value pairs. Amazon DynamoDB is a serverless NoSQL key-value and document database: no servers to manage, single-digit millisecond responses at any scale, and capacity that grows with traffic. It suits shopping carts, sessions and anything with huge, spiky traffic and simple lookups by key.

Exam tip Key-value data, massive scale, serverless, millisecond speed = DynamoDB. Complex joins across many tables = relational (RDS or Aurora).

Memory-based: Amazon ElastiCache

Reading from memory is far faster than reading from disk. Amazon ElastiCache is a managed in-memory data store and cache, compatible with Valkey, Memcached and Redis OSS. Put it in front of the database so the busiest product pages come from memory in microseconds, and the database does less work.

Dev: The squeaky hedgehog page gets read a million times. Cache it once, serve it from memory.

The whole picker

Three more specialists finish the set. Amazon Redshift is a data warehouse: SQL analytics over years of history, not the shop's live orders. Amazon Neptune is a graph database for data that is all relationships ("customers who bought this also bought"). Amazon DocumentDB stores JSON (JavaScript Object Notation) documents and works with MongoDB apps. Start from the data's shape, then pick.

AWS DMS and AWS SCT

AWS Database Migration Service (AWS DMS) moves data into AWS while the source database stays fully usable, and keeps the copy in sync with continuous replication until you cut over. It handles same-engine moves (Oracle to Oracle) and moves between engines (Oracle to Aurora). For a change of engine, the AWS Schema Conversion Tool (AWS SCT) first converts the database's structure (schema) and code to the new engine's format.

Ranger: carries his food bowl to a new spot, one kibble at a time

Sam: That's replication.

Exam tip Same engine: DMS alone. Different engines: SCT converts the schema, then DMS moves the data. "Minimal downtime" = DMS replication.

Object storage: Amazon S3

Object storage keeps each file as an object (the data plus metadata) in a flat container. Amazon Simple Storage Service (Amazon S3) stores objects in buckets, with practically unlimited space and very high durability (designed for 99.999999999%, "11 nines"). Uses: product photos, backups, logs, data lakes for analytics and static websites. You read and write whole objects over the web, not individual disk blocks.

Rosa: So every product photo is an object, and the bucket is the photo album?

Dev: A photo album with no last page.

Block storage: Amazon EBS and instance store

Block storage acts like a hard drive for a server. Amazon Elastic Block Store (Amazon EBS) volumes attach to EC2 instances in one AZ and keep their data when the instance stops; snapshots back them up to S3. An instance store is disk physically on the host: very fast but temporary, so its data is lost when the instance stops or terminates. Use it for caches and scratch data only.

Exam tip Must survive a stop? EBS. Fast scratch space you can lose? Instance store.

File storage: Amazon EFS and Amazon FSx

File storage gives many servers a shared folder at the same time. Amazon Elastic File System (Amazon EFS) is a serverless file system for Linux, using the Network File System (NFS) protocol; it grows and shrinks automatically and can be shared across AZs. Amazon FSx runs popular file systems as managed services, such as FSx for Windows File Server (with Server Message Block (SMB) and Active Directory), FSx for NetApp ONTAP and FSx for OpenZFS.

Exam tip Shared Linux files = EFS. Windows file shares = FSx for Windows File Server.

The storage-class ladder

Every S3 storage class trades storage price against access. S3 Standard is for data read often. S3 Intelligent-Tiering moves objects between tiers for you when access is unknown or changing. S3 Standard-IA (Infrequent Access) and S3 One Zone-IA (one AZ only, for data you can re-create) cost less to store but charge to read. The S3 Glacier classes are for archives: Instant Retrieval (milliseconds), Flexible Retrieval (minutes to hours) and Deep Archive (hours, the lowest cost).

Exam tip Down the ladder: cheaper storage, higher retrieval fees. Only Glacier Flexible Retrieval and Deep Archive make you wait; the rest read in milliseconds. Unknown pattern = Intelligent-Tiering.

Lifecycle policies do the moving

A lifecycle policy (lifecycle rule) tells S3 to act on objects by age. Transition actions move objects to a cheaper class, such as Standard-IA after 30 days and Glacier Deep Archive after a year. Expiration actions delete objects you no longer need, such as old logs after 400 days. Set it once and it runs on every matching object.

Mr. Barkley: So 2019 retires to the cheap seats by itself. Good.

Exam tip Use cases: data that cools with age, logs with a retention period, compliance archives. Watch minimum storage durations on IA and Glacier classes.

AWS Storage Gateway: a cache in the warehouse

AWS Storage Gateway connects on-premises apps to cloud storage. It runs in your building and keeps a local cache of the most-used data, so access feels fast, while the full data set lives in AWS. Types include a file gateway (file shares backed by S3), a volume gateway (block volumes) and a tape gateway (virtual tapes for existing backup software). The exam calls these cached file systems.

Rosa: So my label server keeps a fast local copy, and the real files sit safely in AWS?

Dev: Exactly.

AWS Backup: one place for every backup

AWS Backup centrally manages and automates backups across AWS services, such as EBS volumes, EC2 instances, RDS and Aurora databases, DynamoDB tables, EFS, FSx and S3. A backup plan sets the schedule and how long copies are kept, and backups can be copied to another Region or account. Use cases: one policy for the whole company, proof for auditors that backups ran, and recovery after mistakes or ransomware.

Exam tip Many services, one backup policy and report = AWS Backup. A single EBS snapshot by hand is not the same thing.

Managed databases

AWS runs the database software, patching, backups and high availability, so you focus on your data.

• Amazon RDS: managed relational databases (MySQL, PostgreSQL, MariaDB, Oracle, SQL Server, Db2)
• Amazon Aurora: AWS's MySQL/PostgreSQL-compatible relational database, built for the cloud
• Amazon DynamoDB: serverless NoSQL key-value database
• Amazon ElastiCache and Amazon MemoryDB: in-memory data stores (Redis OSS/Valkey, Memcached)
• Amazon DocumentDB (MongoDB-compatible), Amazon Neptune (graph), Amazon Keyspaces (Cassandra)
• Amazon Redshift: data warehouse for analytics

Managed vs self-managed

You can also install a database on EC2 yourself, but then you do the patching, backups and failover. Exam questions that say 'reduce operational overhead' point to the managed service.

Relational = tables and SQL

Relational databases store data in tables with fixed columns and relationships, queried with SQL. Good for transactions, orders and accounts.

RDS and Aurora

Amazon RDS runs familiar engines and handles backups, patching and Multi-AZ standby copies for high availability; read replicas add read capacity. Amazon Aurora is compatible with MySQL and PostgreSQL but faster and more resilient, storing six copies of data across three AZs, with Aurora Serverless for variable workloads.

Exam tip Multi-AZ = availability (automatic failover). Read replicas = read performance.

What NoSQL means

NoSQL databases don't need fixed tables. Amazon DynamoDB is a key-value and document database: each item has a key and flexible attributes.

• Serverless: no servers or versions to manage
• Single-digit millisecond performance at any scale
• On-demand or provisioned capacity
• Global tables replicate across Regions
• DynamoDB Accelerator (DAX) adds an in-memory cache

When to choose it

High-traffic apps with simple lookups (shopping carts, game sessions, user profiles), unpredictable scale, and no need for complex joins.

Exam tip Complex joins and transactions across tables = relational (RDS/Aurora). Massive scale, simple key lookups = DynamoDB.

What it does

Amazon ElastiCache keeps data in memory (Redis OSS, Valkey or Memcached engines) so reads take microseconds. It usually sits in front of another database as a cache: frequently read data is served from memory, taking load off the database.

• Session stores and leaderboards
• Caching database query results
• Real-time counters

Remember

Memory is fast but more expensive and (for a pure cache) temporary; the main copy of the data stays in the database. Amazon MemoryDB is the durable in-memory option when memory is the primary database.

Exam tip 'Improve read performance of a database with repeated queries' = ElastiCache.

AWS DMS

AWS Database Migration Service copies data from a source database to a target with the source staying online. It supports homogeneous migrations (Oracle to Oracle) and heterogeneous ones (Oracle to Aurora), and continuous replication to keep them in sync until cut-over.

AWS SCT

When the engines differ, the schema (tables, views, stored procedures) must be converted first. The AWS Schema Conversion Tool (and the newer DMS Schema Conversion in the console) converts the schema and flags code it can't convert automatically.

Exam tip Same engine: DMS only. Different engines: SCT (or DMS Schema Conversion) for the schema, then DMS for the data.

Three kinds of storage

Storage comes in three shapes, each for a different job:

• Object storage: Amazon S3 (files as objects in buckets, unlimited, accessed over HTTPS)
• Block storage: Amazon EBS volumes and instance store (disks for EC2)
• File storage: Amazon EFS (Linux, NFS) and Amazon FSx (Windows File Server, Lustre, NetApp ONTAP, OpenZFS)

Also in the list

Amazon S3 Glacier storage classes for archives; AWS Storage Gateway (hybrid access to cloud storage); AWS Backup (central backup policies across services); AWS Elastic Disaster Recovery; the AWS Snow Family for moving data offline.

Amazon EBS

EBS volumes are network-attached disks for EC2 in one Availability Zone. Data persists when the instance stops. You can resize, change type (SSD gp3/io2, HDD), and take snapshots (stored in S3, copyable to other Regions).

Instance store

Instance store is a disk physically attached to the host. Very fast, included in the instance price, but temporary: the data is lost when the instance stops, terminates or the hardware fails. Use it for caches, buffers and scratch data.

Exam tip Data must survive a stop: EBS. Temporary, fastest local scratch space: instance store.

Amazon EFS

Amazon EFS is a managed NFS file system for Linux that many EC2 instances (and containers, Lambda) can mount at the same time. It grows and shrinks automatically and spans multiple AZs. Infrequent Access and Archive classes lower cost.

Amazon FSx

Amazon FSx gives you popular file systems fully managed: FSx for Windows File Server (SMB, Active Directory integration), FSx for Lustre (high-performance computing), FSx for NetApp ONTAP and FSx for OpenZFS.

Exam tip Shared storage for Windows servers: FSx for Windows File Server. Shared storage for Linux instances: EFS.

What it is

AWS Storage Gateway is a hybrid service: a virtual or hardware appliance in your data center that gives local apps standard file, volume or tape access, while storing the data in AWS. Frequently used data is cached locally for low-latency access.

• S3 File Gateway: NFS/SMB file shares backed by S3
• FSx File Gateway: local cache for FSx for Windows File Server (no longer offered to new customers)
• Volume Gateway: iSCSI block volumes backed up to AWS
• Tape Gateway: replaces physical backup tapes with virtual tapes in S3 Glacier

Exam pattern

'On-premises servers need low-latency access to data stored in S3' or 'replace our backup tapes' = Storage Gateway.

Archive classes

S3 Glacier storage classes keep data you rarely need at very low cost:

• Glacier Instant Retrieval: milliseconds; data read about once a quarter
• Glacier Flexible Retrieval: expedited (1-5 minutes), standard (3-5 hours) or bulk (5-12 hours)
• Glacier Deep Archive: lowest cost; standard retrieval within 12 hours (bulk up to 48)

Using them

Put data there with S3 Lifecycle rules or directly. There are minimum storage durations (90 days, 180 days for Deep Archive) and retrieval charges. S3 Glacier Vault Lock / S3 Object Lock can make archives write-once-read-many for compliance.

Exam tip Regulatory records kept for years, rarely read, retrieval can wait a day = Glacier Deep Archive.