What the exam asks
- Choose an instance family and size from workload traits, and recognise burstable instances running out of credits.
- Choose cluster, spread or partition placement groups, plus ENA or EFA networking.
- Pick AWS Batch, Amazon EMR, Lambda or Fargate for batch, big-data and event workloads, and size Lambda memory to gain CPU.
- Place compute close to users or data with Local Zones, Wavelength or Outposts.
- Decouple tiers so that each one scales on its own metric.
Core ideas
Instance families
| Family | Letters | Optimised for | Exam trigger |
|---|---|---|---|
| General purpose | M, T (burstable), Mac | Balanced CPU and memory | Web and app servers, small databases, dev/test |
| Compute optimised | C | High CPU per GiB of memory | Batch processing, media transcoding, HPC, game servers, ML inference, ad serving |
| Memory optimised | R, X, z, U (High Memory) | Large RAM per vCPU | In-memory databases and caches, real-time big-data analytics, SAP HANA |
| Storage optimised | I, D, H | Local NVMe or HDD with very high sequential or random I/O | NoSQL databases, data warehousing, distributed file systems |
| Accelerated computing | P, G, Trn, Inf, F | GPUs, AWS ML chips, FPGAs | ML training (P, Trn), inference (Inf, G), graphics rendering (G) |
Suffix letters matter too: g = AWS Graviton (Arm, best price performance for code that is portable to Arm), a = AMD, i = Intel, d = local NVMe instance store, n = enhanced networking bandwidth.
Burstable T instances earn CPU credits below their baseline and spend them above it. When CPUCreditBalance hits zero in standard mode, the CPU is throttled to baseline. Unlimited mode avoids the throttling but charges for surplus credits. Workloads that run belong on a fixed-performance family (M or C), which costs less than paying for surplus credits.