Amazon EC2 (Elastic Compute Cloud)

Amazon EC2 provides resizable virtual machines (instances) in the AWS cloud. It remains the foundational AWS compute primitive — even managed services like RDS, EMR, and SageMaker run on EC2 underneath. EC2 is the right choice when you need full OS-level control, specialized hardware (GPU, high-memory, HPC), or software that doesn't fit containers or Lambda.

Key Concepts:

• Instance Families: General purpose (M, T), compute-optimized (C), memory-optimized (R, X), storage-optimized (I, D), accelerated (G, P, Trn, Inf), and burstable (T).
• AMIs (Amazon Machine Images): Immutable OS + software snapshots used to launch instances; build with EC2 Image Builder or packer.
• EBS vs. Instance Store: EBS is network-attached, durable, and snapshotable; instance store is local NVMe, ephemeral, and fastest — useful for scratch space.
• Placement Groups: Cluster (low-latency), Spread (AZ-isolated), Partition (rack-isolated) — control where instances land relative to each other.
• Auto Scaling Groups: Launch/terminate instances based on metrics, schedules, or health checks; integrate with ELB for zero-downtime deploys.
• Nitro System: Hypervisor + hardware offload that powers modern instance families — enables bare-metal instances and strong tenant isolation.

Pricing Models:

• On-Demand: Per-second billing, no commitment — simple but most expensive.
• Savings Plans / Reserved Instances: 1- or 3-year commitment for up to ~72% discount. Savings Plans apply across instance families; RIs are instance-specific.
• Spot Instances: Up to ~90% off on-demand pricing using spare capacity — may be interrupted with 2 minutes' notice. Ideal for batch, CI, big data, ML training with checkpointing.
• Dedicated Hosts / Dedicated Instances: Physical server isolation for BYOL licensing and compliance.
• Capacity Reservations: Guarantee capacity in a specific AZ without commitment pricing — useful for DR and predictable scale events.

When to Use EC2 (vs. Alternatives):

• Pick Lambda for short-running, event-driven work under 15 minutes.
• Pick Fargate / ECS / EKS for containerized workloads.
• Pick Batch / EMR for large-scale parallel jobs that AWS can orchestrate for you.
• Pick EC2 when you need OS-level control, GPUs, specialized networking (EFA), custom kernels, or long-lived stateful processes.

Best Practices:

• Immutable infrastructure — bake AMIs, don't configure live.
• Use Systems Manager Session Manager instead of opening SSH (port 22) to the internet.
• Attach instance roles rather than embedding AWS credentials.
• Distribute across AZs behind an Elastic Load Balancer for HA.
• Tag aggressively — cost allocation, automation, and access control all rely on tags.