Compute Services (ECS)
Core concepts
Elastic Compute Service (ECS) is the flagship IaaS product. A candidate must be fluent in the instance family taxonomy: general purpose (g), compute optimized (c), memory optimized (r), and shared/burstable types. Match the family to the workload - web servers to general purpose, batch encoding to compute optimized, in-memory caches to memory optimized.
Know the image types (public, custom, shared, marketplace), disk categories (ESSD, SSD, ultra disk, basic), and network options (classic vs VPC - VPC is the modern default). Understand Security Groups as stateful virtual firewalls attached to instances, with allow-only rules and default deny inbound.
Auto Scaling groups instances behind a scaling policy and pairs naturally with SLB. User data bootstraps configuration at first launch. Snapshots back up disks; they are incremental and stored in OSS-backed infrastructure.
Traps
- Security groups are allow-only and stateful; network ACLs are stateless and can deny.
- Resizing an instance changes CPU/memory but not necessarily disk; you may need to resize the disk separately.
- Stopping an instance does not delete it, but billing behavior differs by network charge type.
Worked example
A team needs a web tier that survives a zone failure and scales with traffic. Answer: an Auto Scaling group spanning multiple zones in one VPC, instances in a general-purpose family, behind an SLB, with a scaling policy on CPU. Snapshots for the data disk, custom image for fast bootstrapping.