Check a fleet against cluster, spread and partition placement group limits and see the blast radius each strategy gives.
The three strategies trade latency against correlated failure. Cluster packs everything into one AZ on the same network spine for the lowest latency and the worst blast radius; spread puts every instance on distinct underlying hardware but caps at 7 per AZ per group, so larger fleets need several groups; partition groups sit in between, giving up to 7 named fault domains per AZ that a partition-aware application like HDFS or Cassandra can map replicas onto. The spread limit of 7 per AZ is the constraint people discover at launch time with an InsufficientCapacity-style error, and blast radius is the number that decides whether one rack failure is an incident or an outage.
EC2 Placement Group
spread capacity = 7 × Availability Zones per group; partition fault domains = partitions × Availability Zones; blast radius = ceil(instances ÷ fault domains) ÷ instances.
spread capacity = 7 × Availability Zones per group; partition fault domains = partitions × Availability Zones; blast radius = ceil(instances ÷ fault domains) ÷ instances. The three strategies trade latency against correlated failure. Cluster packs everything into one AZ on the same network spine for the lowest latency and the worst blast radius; spread puts every instance on distinct underlying hardware but caps at 7 per AZ per group, so larger fleets need several groups; partition groups sit in between, giving up to 7 named fault domains per AZ that a partition-aware application like HDFS or Cassandra can map replicas onto.
The spread limit of 7 per AZ is the constraint people discover at launch time with an InsufficientCapacity-style error, and blast radius is the number that decides whether one rack failure is an incident or an outage.
This calculator takes 4 inputs: Placement strategy, Instances to place, Availability Zones available, Partitions per Availability Zone. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
Not inside one group — a placement group has a single strategy and an instance belongs to at most one group. The usual pattern is a partition group for the stateful tier and no placement group at all for stateless instances, which AWS already spreads across the AZs you launch into.