Domain 3 of 4 · Chapter 2 of 5

High-Performing and Elastic Compute

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing AWS Certified Solutions Architect – Associate premium course — no separate purchase.

Included in this chapter:

  • Instance family deep dive: when each family beats the alternatives
  • Graviton compatibility checklist + migration playbook
  • Lambda performance tuning: memory, provisioned concurrency, SnapStart
  • Placement groups + EFA: HPC patterns that need both

EC2 instance families at a glance

FamilyProfileWorkload fitGraviton variant
M (m6i, m7g)Balanced CPU/RAMGeneral-purpose web / app serversm6g / m7g
C (c6i, c7g)Compute-optimized (high CPU/RAM ratio)Batch processing, gaming, scientificc6g / c7g
R (r6i, r7g)Memory-optimizedIn-memory DBs, large JVMs, real-time analyticsr6g / r7g
X (x2idn, x2iedn)Extra-large memorySAP HANA, large Apache Spark—
I (i4i)High local NVMe IOPSNoSQL DBs (Cassandra, MongoDB, etc.)—
D (d3)Dense HDD storageData warehousing, MapReduce—
P (p5)GPU (NVIDIA H100)ML training, HPC—
G (g5)GPU (cheaper than P)ML inference, graphics rendering—
Inf2 / Trn1nML accelerator (AWS chips)ML inference (Inf2) / training (Trn1n)ARM-based
T (t3, t4g)Burstable CPU creditsCheap, bursty workloadst4g
Hpc (Hpc6id, Hpc7a)HPC + EFA networkingTightly-coupled HPC clusters—
Cheat sheet

Cheat sheet

  • Match the instance family to the workload's bottleneck resource
  • On the compute spectrum, pick the most-managed option that still meets your control needs
  • Exceeding any Lambda ceiling (15 min / 10 GB RAM / 10 GB /tmp) → move to Fargate, Batch, or EC2
  • Latency-sensitive Lambda → provisioned concurrency to eliminate cold starts
  • Choose the placement group by goal: Cluster for latency, Spread for isolation, Partition for distributed stores
  • Tightly-coupled HPC / multi-node ML training → add an Elastic Fabric Adapter
  • Sustained-load workload → don't run it on a burstable T-family instance
  • Java / Python / .NET Lambda with slow cold starts → enable SnapStart (free for Java)
  • Drain Spot capacity on the rebalance recommendation, not the 2-minute notice
  • In an ECS capacity provider strategy, base guarantees a minimum and weight splits the remainder
  • Tune ECS target-tracking with a short scale-out and long scale-in cooldown
  • Make EKS Cluster Autoscaler prefer one node group → priority expander
  • Run setup/cleanup during scaling → ASG lifecycle hooks

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. Amazon EC2 instance types
  2. Burstable performance instances (T family)
  3. Elastic Fabric Adapter (EFA)
  4. Amazon EC2 documentation (root)
  5. Amazon ECS overview
  6. Amazon EKS overview
  7. AWS Fargate
  8. AWS Lambda Developer Guide
  9. AWS Graviton processor
  10. Lambda quotas (15-min, memory, /tmp limits)
  11. Lambda provisioned concurrency
  12. Placement groups
  13. Lambda SnapStart
  14. https://aws.amazon.com/lambda/pricing/
  15. Spot Instance interruptions
  16. https://docs.aws.amazon.com/autoscaling/ec2/userguide/lifecycle-hooks.html
  17. https://aws.amazon.com/ec2/instance-types/x2i/
  18. https://aws.amazon.com/ec2/instance-types/u7i/
  19. https://aws.amazon.com/hpc/efa/