cluster (high performance): deploy to same rack (and same AZ). High performance + High risk. Good for HPC. 10 Gbps network
spread (high availability): deploy to different racks (at most one instance per rack) across different AZ's (in the same region). Limited to 7 instances per AZ
partition (balanced): deploy to the same rack (and same AZ), but they are replicated to other racks (and other AZs) as a backup. The number of partitions define the number of replicas. Use case for HDFS, HBase, Cassandra, Kafka