AWS Database Choices, Use Cases and Characteristics

I’m currently studying for my AWS Certified Solutions Architect Pro exam and one of the main topics I always seem to have something of a blind spot about it databases. What I’ve done below is to try and summarise some points about RDS and the other database offerings on AWS as a quick reference guide for the exam. If it helps you either for the exam or in real life, great!

The notes were put together using AWS documentation and Re:Invent 2015 videos. If you spot any factual errors, please tweet me @ChrisBeckett and I’ll correct them.

Key differences between SQL and NoSQL

• NoSQL is schema-less, easy reads and writes, simple data model
• NoSQL scaling is easy
• NoSQL focusses on performance and availability at any scale
• SQL has a strong schema, complex relationships, transactions and joins
• SQL scaling is difficult
• SQL focusses on data consistency over scale and availability

What is DynamoDB?

• NoSQL database offering
• Fully managed by AWS (no need for separate EC2 instances, etc)
• Single digit millisecond latency
• Massive and seamless scalability at low cost

Use cases for DynamoDB

• Internet of Things (Tracking data, real time notifications, high volumes of data)
• Ad Tech (ad serving, ID lookup, session tracking)
• Gaming (gaming leader boards, usage history, logs)
• Mobile and Web (Storing user profiles, session data)
• All above use cases require high performance, high scale and high volume

DynamoDB Characteristics

• Automatically replicated (writes) across three Availability Zones in a single region, persisted to SSD. Write is confirmed when master copy and one replica is updated
• Reads can be eventually or strongly consistent, no latency trade off, strongly consistent data read from master only
• DynamoDB consists of tables, tables have items, items have attributes. Because there is no schema, hash keys or item IDs must be present to identify an entry in a table
• In order to scale properly, you just tell DynamoDB what throughput you need and it will configure the infrastructure for you
• Pay for the amount of storage used and the amount of throughput (reads and writes)
• Free tier entitlement of 25GB and 60 million reads and 60 million writes per month. Very cost effective

What is RDS?

• Fully managed relational databases (MySQL, PostgreSQL, SQL Server, Oracle, Aurora)
• Fast, predictable performance
• Simple and fast to scale
• Low cost, pay for what you use

Use cases for RDS

• Anything that requires a SQL back end, such as existing corporate applications
• RDS supports VPC, high availability, instance scaling, encryption and read replicas for Aurora, MySQL, PostgreSQL
• MySQL even supports cross region deployment
• 6TB max storage limit for Oracle, PostgreSQL and MySQL. Aurora is 64TB and SQL Server 4TB
• Scale storage for all except SQL Server
• Provisioned IOPS 30,000 for MySQL, PostgreSQL, Oracle. 20,000 for SQL Server
• Largest RDS instance supported is R3.8XL for all platforms

What is Aurora?

• 99.99% availability
• 5x faster than MySQL on the same hardware
• Distributed storage layer, so scales better
• Data replicated six times across three availability zones
• 15 read replicas maximum
• Fully MySQL compatible

RDS Characteristics

• Supports three types of storage
  • General purpose SSD for most use cases
  • Provisioned IOPS for guaranteed storage performance of up to 30,000 IOPS
  • Magnetic for inexpensive very small workloads
• Multi-AZ support – provides automatic failover, synchronous replication and is inexpensive and simple to set up
• Can create read replicas in other regions, can promote to a master for easy DR or put data close to the users that use them
• Pay for what you consume, some free tier entitlements exist for RDS (20GB of data storage, 20GB of backups, 10m IOPS, 750 micro DB instance hours)

What is ElastiCache?

• In memory key value store
• High performance
• Choice of Redis or Memcached
• Fully managed, zero admin

ElastiCache Use Cases

• Caching layer for performance or cost optimisation of an underlying database
• Storage of ephemeral key-value data
• High performance application patterns such as session management, event counters, etc
• Memcached is the simpler of the two. Cache node auto discovery and multi-AZ node placement
• Redis is Multi-AZ with auto failover, persistence and read replicas
• Redis handles more complex data types
• Monthly bill is number of nodes * duration nodes were used for
• Some free tier eligibility – 750 micro cache node hours

What is Amazon Redshift?

• Relational data warehouse
• Massively parallel, petabyte scale
• Fully managed

Redshift Use Cases

• Leverage BI tools, Hadoop, machine learning, streaming
• Pay as you go, grow as you need
• Analysis in line with data flows
• Managed availability and data recovery

Redshift Characteristics

 

• HDD and SSD platforms
• Architecture has a leader node (endpoint for communication) and compute nodes. These are linked by 10Gbps networking
• Leader node also stores metadata and optimises the query plan
• Compute nodes have local columnar storage and have distributed, parallel execution of queries, backups, loads, restores, resizes
• Backed up continuously and incrementally to S3 and across regions
• Streamed restores
• Tolerates disk failures, node failures, network failures and AZ/region failures
• Leader node is free, compute nodes are charged as they are used