Amazon EC2

What it is
- Elastic Compute Cloud — resizable virtual machines (instances) in AWS. The baseline compute service everything else abstracts away.
- You control the OS, networking, storage, and everything on the instance; AWS manages the physical host.
Pay only for what you run
- Billed per second (Linux) or per hour, only while the instance runs. Stop it and compute charges stop.
- Commercial terms are a separate axis — see Instance Pricing Options (On-Demand, Reserved, Spot, etc.).

Anatomy of an instance

Amazon Machine Image (AMI) — the template: OS, software, and config the instance boots from.
Instance type — the hardware profile (vCPU, memory, network, GPU). Format family.generation.size, e.g. t3.micro.
Storage — Amazon Elastic Block Store (EBS) volumes (persistent, network-attached) and/or EC2 Instance store (ephemeral, physically attached, wiped on stop).
Networking — lives in an Amazon VPC subnet; a security group is the instance’s stateful firewall.
Key pair — SSH/RDP credentials (or skip keys entirely with Session Manager).
User data — a bootstrap script run once at first launch to install/configure software.

Chosen by the bottleneck of your workload:

General Purpose (T3) — balanced CPU/memory. Web servers, dev, small DBs. T-family is burstable (CPU credits).
Compute Optimized (C5) — high CPU per dollar. Batch, gaming, media transcoding, HPC.
Memory Optimized (R5a) — large RAM. In-memory caches, high-performance and in-memory databases.
Storage Optimized (D2) — high sequential disk throughput / IOPS. Data warehouses, distributed file systems.
Accelerated Computing (F1) — hardware accelerators (GPU, FPGA). ML training/inference, rendering.

Running → Stopped — EBS-backed instance shuts down, EBS data persists, no compute charge; instance-store data is lost.
Stopped → Started — usually lands on different host hardware; public IP changes unless an Elastic IP is attached.
Terminated — instance deleted; root EBS volume deleted by default.
Reboot — same host, IPs and instance-store data preserved.

EC2 Auto Scaling — an Auto Scaling Group adds/removes instances to match demand and replaces unhealthy ones. Horizontal scaling.
Vertical scaling = resize to a bigger instance type; requires a stop/start.
Spread instances across multiple Availability Zones behind a load balancer for high availability.
Amazon EC2 Fleet — provision a mixed group across instance types, AZs, and purchase options in one API call.

AWS — the hypervisor, host hardware, physical security, network infrastructure.
You — guest OS patching, software, security group rules, IAM, and data on the instance.

Amazon Lightsail — bundled EC2 at a fixed monthly price, knobs hidden; EC2 is the full-control version.
AWS Lambda / AWS Fargate — serverless compute with no instance to manage; EC2 is you managing the server.