AWS Knowledge Hub

English

Tiếng Việt

English

Tiếng Việt

Appearance

EC2 Placement Groups ​

Summary ​

  • A placement group controls how EC2 instances are placed on underlying hardware to optimize either network performance or availability.
  • There are three types: Cluster (low‑latency, high throughput within one AZ), Spread (maximum instance‑level isolation), and Partition (group‑level isolation for large distributed systems).
  • Choosing the right placement group can significantly improve performance characteristics or reduce correlated failure risk without changing your application logic.

Placement group comparison ​

mermaid
mindmap
  root((EC2 Placement Groups))
    Cluster
      Same rack / AZ
      Low latency, high throughput
      HPC, tightly coupled apps
    Spread
      Different racks
      Max isolation (7 instances/AZ)
      Critical apps
    Partition
      Multiple partitions per AZ
      Distributed systems
      Hadoop, Cassandra, Kafka

Best Practices ​

  • Plan placement groups up front: you cannot change the type later, so decide Cluster/Spread/Partition during architecture design, not after deployment.
  • Cluster placement group: use for HPC, ML training, and tightly coupled workloads that require very low latency and high throughput; combine with enhanced networking and compatible instance types.
  • Spread placement group: use for a small number of critical instances that must not share hardware and need maximum isolation (up to 7 per AZ).
  • Partition placement group: use for big distributed systems where failures should be isolated at the partition level (Hadoop, Cassandra, Kafka); design the application to be partition‑aware.
  • Always monitor network metrics (latency, throughput, packet loss) and status checks; placement groups do not remove the need for resilient application design.
  • Remember that placement groups are free, but the wrong type can increase risk (e.g. single‑AZ Cluster) or limit scale (Spread instance cap).

Exam Notes ​

  • Many questions ask you to pick the correct placement group type for a scenario:
    • Tight, low‑latency communication within one AZ → Cluster.
    • A few critical instances needing maximum hardware isolation → Spread.
    • Hundreds of nodes in a distributed store/analytics cluster → Partition.
  • Memorize key limits: Spread ≤ 7 instances per AZ, Partition ≤ 7 partitions per AZ, and Cluster is single‑AZ only.
  • Placement groups do not replace Multi‑AZ or Multi‑Region DR; Cluster can actually increase blast radius if the AZ fails.
  • Capacity/instance‑type constraints can prevent launches into a placement group; that is a common pitfall in exam scenarios.

AWS documentation ​