AWS Deep Dive
- author:: Nathan Acks
- date:: 2022-07-05
Today I’ll be starting the ~6 hour “AWS Cloud Practitioner Essentials” course as I continue to dive deeper into AWS.
Introduction to Amazon Web Services
Wait… AWS has per-minute satellite rental?!?
Compute in the Cloud
- General Purpose
- Compute Optimized
- Memory Optimized
- Accelerated Computing
- Storage Optimized
“Accelerated computing” here means “hardware accelerated computing”. So, high-end GPUs, cryptographic accelerators, custom AI/ML chips, etc.
“Storage optimized” instances aren’t necessarily designed to support a lot of local storage (though the can, and often do, this), but are rather designed to support for high data throughput for that local storage.
On-Demand (the default model)
Savings Plan (commit to a certain numbers of hours of compute time; can include services like Lambda, and usage above the agreed-upon amount is charged at the regular rate)
Reserved Instances (commit to a certain number of instances)
Spot Instances (additional capacity purchased on the spot market; can be reclaimed by AWS with a 2 minute warning)
Dedicate Hosts (single-account hosts, mostly for high security/compliance workloads)
Both the “savings plan” and “reserved instances” offer similar savings, but over different metrics.
Elastic Load Balancers
Elastic Load Balancing (ELB) handles both load balancing between regions (based on regional capacity and the proximity of a request to a particular region) and within regions (by distributing load uniformly between all instances in a region).
ELB doesn’t just have to be client facing, however - it can also be used between the front-end and back-end systems in an architecture.
Messaging and Queueing
Simple Queue Service (SQS) - Queue/Buffer messages from one service to another within AWS
Simple Notification Service (SNS) - Queue/Buffer messages from a service in AWS to another service that may be inside of or outside of AWS (and could be directly user-facing, like SMS or push notifications)
SQS is a pure machine-to-machine buffer, while SNS is based around a one-to-many pub/sub model.
Additional Compute Services
Amazon’s two container management services are ECS (Elastic Container Service) for Docker and EKS (Elastic Kubernetes Service) for Kubernetes. These must be paired with an underlying platform where the containers will actually run - either EC2 (for traditional containerization) or “Fargate” (which functions in a “serverless” fashion like Lambda).