A company will deploy a distributed in-memory database across EC2 instances: one primary node and eight worker nodes. The primary node handles cluster health, accepts client requests, distributes work to worker nodes, and aggregates responses. Worker nodes replicate data partitions among themselves. The application requires the lowest possible network latency for maximum performance. Which placement and instance choice is best?

The correct choice is to launch memory-optimized EC2 instances in a cluster placement group. Memory-optimized instances are ideal for in-memory databases due to their high RAM-to-vCPU ratio. A cluster placement group ensures that all instances are launched within a single Availability Zone on hardware that supports high-bandwidth, low-latency network connectivity, which is critical for the distributed database's performance and inter-node communication. Partition placement groups segment instances into logical partitions to reduce correlated failures, but they don't guarantee the lowest latency within the group. Spread placement groups place instances on distinct underlying hardware to minimize correlated failures, but this can increase latency. Compute-optimized instances are not the best fit for an in-memory database, which primarily requires high memory capacity and throughput.