Design High-Performing Architectures
This domain accounts for 24% of the AWS SAA-C03 exam — the largest single domain. It tests your ability to select the right AWS service and configuration for a given performance, scalability, or throughput requirement. Expect scenario-based questions where two or three answers are plausible; winning requires knowing exact limits, default behaviors, and the specific capability that distinguishes one service from another.
EC2 Instance Families, Placement Groups, and EBS Volume Types
EC2 Instance Families
| Family | Optimized For | Exam Keywords |
|---|---|---|
| M (general purpose) | Balanced CPU/RAM | Web servers, small DBs |
| C (compute optimized) | High CPU, low RAM cost | HPC, batch, gaming servers |
| R / X / z (memory optimized) | Large in-memory datasets | In-memory caches, SAP HANA |
| I / D / H (storage optimized) | High sequential disk I/O | OLTP, data warehouses, Hadoop |
| P / G / Inf / Trn (accelerated) | GPU / ML inference | Deep learning, video encoding |
| T (burstable) | Baseline + burst CPU credits | Dev/test, low-traffic web |
Placement Groups
| Type | Behavior | Use Case | Limitation |
|---|---|---|---|
| Cluster | Same rack, same AZ | Ultra-low latency, HPC | Single AZ; failure domain is one rack |
| Spread | Different racks, different AZs | HA critical instances | Max 7 instances per AZ per group |
| Partition | Groups of instances on separate racks | Large distributed systems (Kafka, HDFS) | Up to 7 partitions per AZ |
EBS Volume Types
| Type | Use Case | Max IOPS | Max Throughput | Key Detail |
|---|---|---|---|---|
| gp3 | General purpose (default) | 16,000 | 1,000 MB/s | IOPS and throughput independently configurable |
| gp2 | General purpose (legacy) | 16,000 | 250 MB/s | IOPS tied to size (3 IOPS/GB) |
| io2 Block Express | Mission-critical OLTP | 256,000 | 4,000 MB/s | Multi-attach capable; 99.999% durability |
| io1 | High IOPS OLTP | 64,000 | 1,000 MB/s | Multi-attach capable |
| st1 | Throughput-heavy sequential | 500 | 500 MB/s | Cannot be boot volume |
| sc1 | Infrequent cold access | 250 | 250 MB/s | Lowest cost block storage |
Common traps:
- Exam says "needs more than 64,000 IOPS" → only io2 Block Express qualifies.
- gp2 IOPS scale with volume size; gp3 IOPS are independent — never confuse them.
- st1/sc1 cannot be used as boot volumes.
- Cluster placement groups require enhanced networking (ENA) enabled instances to realize low-latency benefits.
Compute Choices: EC2 vs Lambda vs Fargate vs ECS/EKS
Decision tree for compute selection:
This flowchart maps a workload requirement to the correct AWS compute service.
| Service | Max Duration | State | Cold Start | Pricing Unit |
|---|---|---|---|---|
| Lambda | 15 minutes | Stateless | Yes (ms to sec) | Per request + GB-second |
| Fargate | Unlimited | Container-scoped | Seconds | vCPU + memory per second |
| ECS on EC2 | Unlimited | Container or volume | None (EC2 running) | EC2 instance hours |
| EKS | Unlimited | Pod/volume | None | EC2 or Fargate |
Common traps:
- Lambda max timeout is 15 minutes — any workload exceeding this needs a different compute.
- Fargate removes EC2 cluster management; ECS/EKS on EC2 retains it.
- Lambda scales to thousands of concurrent executions automatically; EC2 Auto Scaling has a warm-up period.
- EKS is Kubernetes-compatible; ECS is AWS-native. Both can run on Fargate.
Worked scenario: A company needs to process image thumbnails triggered by S3 uploads, each taking under 10 seconds. → Lambda with an S3 event trigger. Cost-optimal, no server management, perfectly event-driven.
Unlock with a paid plan
10 more sections (about 2,721 words) continue below. Every paid plan unlocks all SAA-C03 study guides, the full question bank and mock exams, from $4.97/month.