What the exam asks
- Rightsize with Compute Optimizer and move to Graviton.
- Choose Lambda, Fargate or EC2 from the utilisation profile.
- Cut non-production cost with schedules, hibernation and deliberately lower availability.
- Pick the cheapest load balancer that works (ALB, NLB or GWLB), and scale horizontally.
- Choose edge and hybrid compute that meets the latency need without over-buying.
Core ideas
Rightsizing
- AWS Compute Optimizer uses CloudWatch metrics to recommend better types and sizes for EC2, Auto Scaling groups, EBS, Lambda and ECS on Fargate. Memory is only visible inside the OS, so install the CloudWatch agent for memory-aware recommendations. Detailed monitoring only adds 1-minute hypervisor metrics, and Trusted Advisor’s low-utilization check looks at CPU and network only.
- Rightsize first, then commit. Buying a Savings Plan for oversized instances locks in the waste.
- Burstable T instances suit a low average CPU with short spikes. In unlimited mode, a sustained high CPU load incurs surplus-credit charges, so steady busy workloads belong on fixed-performance families.
Graviton
AWS Graviton (Arm-based) instances, such as the M7g, C7g, R7g and T4g families, offer the best price performance for most general workloads. Graviton is also available for Lambda (arm64), Fargate, RDS, Aurora, ElastiCache, OpenSearch and EMR. Java, Python, Node.js, .NET and Go applications usually move with little or no change. Native x86 binaries need to be recompiled, and container images must be built for Arm or as multi-architecture images.
Lambda vs Fargate vs EC2
| Lambda | Fargate | EC2 | |
|---|---|---|---|
| You pay for | Requests plus GB-seconds of run time | vCPU and memory per second of each task | Instance time, whether busy or idle |
| Idle cost | None | None once tasks scale to zero | Full |
| Sweet spot | Spiky, intermittent, event-driven, short (up to 15 minutes) |