Compute and Application Services
What to know
ECS instances come in families tuned for compute, memory, or general purpose, plus GPU/heterogeneous options. Know instance attributes: vCPU/memory ratio, network bandwidth, and whether the family is burstable. Images define the OS; security groups control instance-level traffic. Auto Scaling adjusts instance count against scaling rules and health checks. For managed application tiers, know Server Load Balancer (SLB) for distributing traffic, and container/serverless options where the provider manages the host. Understand that managed services shift patching and availability of the platform to the provider, while you still configure and secure the application.
Traps
Burstable instances accumulate CPU credits and throttle when exhausted - a classic performance-question trap. Questions may conflate scaling out (add instances) with scaling up (resize). Auto Scaling removes unhealthy instances via health checks; a misconfigured health check causes flapping. Also watch for assuming SLB provides application-layer routing when a Layer 4 listener is specified, or that a managed service removes your responsibility for access control and data.
Rule of thumb
Match instance family to the bottleneck: CPU-bound → compute-optimized; memory-heavy (caches, in-memory DB) → memory-optimized; steady low utilization → burstable. Use Auto Scaling plus SLB for resilient stateless tiers.
Worked example
A web tier sees variable traffic and must self-heal. Put stateless ECS instances behind SLB, attach an Auto Scaling group with a health check, and set scale-out rules on CPU - not a single large instance.