An HPC team is building a tightly coupled MPI simulation cluster and wants the lowest possible inter-instance network latency and the highest throughput between nodes, achieved by grouping all nodes together using a placement group. Which statement about a Cluster placement group is correct?
- All instances in a Cluster placement group must reside in a single Availability Zone
- A Cluster placement group can span multiple Availability Zones to increase resilience
- A Cluster placement group guarantees that no single hardware failure can affect more than one instance
- A Cluster placement group is required before you can enable enhanced networking on an instance
Why A? And why not the others?
Correct answer: A. All instances in a Cluster placement group must reside in a single Availability Zone
A Cluster placement group packs instances close together within a single Availability Zone specifically to minimize inter-instance network latency and maximize throughput for tightly coupled workloads such as HPC and MPI simulations, which means it cannot span multiple Availability Zones at all; spanning multiple AZs to increase resilience instead describes a Spread placement group's design goal. Because a Cluster placement group intentionally packs instances close together on related hardware, it does the opposite of guaranteeing isolation from a single hardware failure — that isolation guarantee describes a Spread placement group, not a Cluster one. Enhanced networking is a capability of supported instance types paired with the right driver, and it can be enabled with or without any placement group; a Cluster placement group is not a prerequisite for it, though combining enhanced-networking-capable instances with a Cluster placement group is a common way to get the full latency and throughput benefit.
Source: AWS EC2 documentation: Placement groups — cluster placement strategy