Compute services

Compute ranges from virtual servers (EC2) to containers (ECS, EKS) to serverless (Lambda, Fargate). Auto Scaling and load balancers make compute elastic and highly available.

Amazon EC2 and instance types

• Amazon EC2: Resizable virtual servers. You choose the instance type, OS, and storage, and you manage the OS.
• General purpose: Balanced compute, memory, and networking (web servers, code repositories).
• Compute optimized: High-performance processors for compute-bound work (batch processing, gaming servers, scientific modeling, HPC).
• Memory optimized: Large memory for big in-memory datasets (in-memory databases, real-time analytics).
• Storage optimized: High, sequential read/write to large local datasets (data warehousing, distributed file systems, high-IOPS transactional databases).
• Accelerated computing: GPUs or other accelerators (machine learning, graphics, video encoding).

Containers

• Amazon ECS: AWS's own fully managed container orchestration service.
• Amazon EKS: Managed Kubernetes. Choose it when you want Kubernetes compatibility.
• Amazon ECR: Managed registry to store container images.

Serverless compute

• AWS Lambda: Run code in response to events without managing servers. Pay per request and compute time. Each invocation can run up to 15 minutes.
• AWS Fargate: Serverless compute engine for containers on ECS or EKS: run containers without managing the EC2 instances underneath.

Other compute

• AWS Elastic Beanstalk: Platform as a service: deploy web apps without managing infrastructure details.
• Amazon Lightsail: Simple virtual private servers with predictable monthly pricing for small sites and apps.
• AWS Batch: Runs batch computing jobs at any scale, provisioning the right compute automatically.
• AWS Outposts: AWS-managed racks of AWS infrastructure installed in your own data center for low-latency or local-data needs.

Elasticity and load balancing

• Auto scaling: Amazon EC2 Auto Scaling (and AWS Auto Scaling for multiple services) adds or removes capacity to match demand, which is how you get elasticity.
• Elastic Load Balancing: Distributes incoming traffic across targets (instances, containers, IPs) in multiple AZs, performs health checks, and removes single points of failure. Types include Application Load Balancer (HTTP/HTTPS) and Network Load Balancer (TCP/UDP, very high performance).