An EC2 Placement Group is a logical grouping of instances within a single Availability Zone that determines their physical placement, influencing network latency, throughput, and fault tolerance based on the chosen strategy: cluster, spread, or partition.
An EC2 Placement Group is a logical grouping that influences how instances are placed on underlying hardware, allowing users to control the proximity of instances for performance, high availability, or fault isolation. Placement groups are essential for applications requiring low latency (HPC, real-time analytics) or high resilience (distributed databases, critical microservices). Once created, you can launch instances into the placement group using the --placement-group parameter. Placement groups cannot be merged, and instances can be moved between groups only by stopping, modifying placement, and restarting.
Cluster Placement Group: Packs instances close together within a single Availability Zone, providing low latency and high network throughput (up to 10 Gbps or more). Ideal for HPC, tightly coupled node-to-node communication, and large-scale simulations. Risk: single point of failure if AZ goes down.
Spread Placement Group: Places instances on distinct underlying hardware across multiple Availability Zones (or within a single AZ). Maximum of 7 running instances per group per AZ. Ideal for critical applications requiring isolation, such as master databases, bastion hosts, and small-scale HA clusters. Provides the highest fault tolerance.
Partition Placement Group: Divides instances into logical partitions, each on a separate rack with independent power and network. Partitions can span multiple AZs. Ideal for large distributed systems (Hadoop, Cassandra, Kafka) where failure of one partition does not affect others. Supports thousands of instances.
Cluster groups should use instances with enhanced networking and placement group awareness (e.g., C5n, M5n, R5n, P3dn) to achieve maximum bandwidth
Spread groups have a strict limit of 7 running instances per AZ; for larger deployments, consider partition groups
Placement groups cannot span multiple VPCs or regions; all instances must be in the same VPC and region
Partition groups have a recommended partition count based on workload; use
describe-placement-groupsto see partition informationYou cannot merge placement groups or move an instance from one group to another without stopping and restarting the instance