AWS Knowledge Hub

English

Tiếng Việt

English

Tiếng Việt

Appearance

EC2 Instance Types ​

Summary ​

  • The instance type controls vCPU, memory, storage, and networking for an EC2 instance, so choosing the right type is essential for both performance and cost.
  • Amazon EC2 provides multiple instance families (General Purpose, Compute Optimized, Memory Optimized, Storage Optimized, Accelerated Computing, etc.) tuned for specific workload characteristics.
  • Each family has several generations and sizes; you should prefer current‑generation types (M6/M7, C6/C7, R6/R7, etc.) for better price/performance and feature support.
  • A good cost strategy combines right‑sizing with the right pricing model (On‑Demand, Reserved, Spot, Savings Plans).

Instance type mindmap ​

mermaid
mindmap
  root((EC2 Instance Types))
    General Purpose
      M family
      T family (burstable)
      Mac instances
    Compute Optimized
      C family
    Memory Optimized
      R family
      X / High Memory
    Storage Optimized
      I family
      D family
      H family
    Accelerated Computing
      P (GPU - training)
      G (GPU - graphics)
      Inf (Inference)
      Trn (Training)
    Pricing
      On-Demand
      Reserved Instances
      Spot
      Savings Plans

Best Practices ​

  • Start from the workload, not the instance name: describe whether it is CPU‑bound, memory‑bound, I/O‑bound, or GPU‑heavy and pick the family accordingly instead of defaulting to a familiar type.
  • Prefer current‑generation families (M6/M7, C6/C7, R6/R7, etc.) to get better performance and usually lower cost per unit of performance than older generations.
  • Use burstable instances (T3/T4g) for light or variable workloads that do not need sustained high CPU; monitor CPU credits to avoid throttling.
  • Separate special workloads: use R*/X*/High Memory for large databases, I*/D*/Hfor big data/analytics, and P/G*/Inf*/Trnfor ML/HPC rather than forcing everything onto M.
  • Right‑size regularly: at least each quarter, review CPU/memory/I/O metrics, downsize or split under‑utilized instances, and consider Spot/Savings Plans for stable portions of usage.
  • Combine instance types with appropriate pricing models: steady workloads → Reserved/Savings Plans; batch/CI/ML experiments → Spot; dev/test → On‑Demand or Spot.

Exam Notes ​

  • Remember the naming pattern: [family][generation].[size] (for example, m6i.large, c7g.xlarge) and map families to primary use cases:
    • General Purpose (M, T, Mac) – balanced workloads.
    • Compute Optimized (C) – CPU‑intensive.
    • Memory Optimized (R, X, High Memory) – in‑memory DB, caches.
    • Storage Optimized (I, D, H) – high throughput/I/O.
    • Accelerated (P, G, Inf, Trn) – GPU/ML/HPC.
  • Exams (especially SAA/DVA) frequently test on burstable instances (T2/T3/T4g and CPU credits), current vs previous generations, instance store vs EBS, and features like EBS‑optimized/enhanced networking.
  • Know the highlights of the Nitro System (better performance, stronger isolation, enhanced networking) and when a metal instance is needed (bare‑metal access, specific licensing).
  • In cost questions, link instance selection to pricing models (for example, c6i with Reserved or Savings Plans for a compute‑intensive, predictable workload).

AWS documentation ​