Cluster

  • Same rack and same availability zone
  • Great network, low latency (10Gbps bandwidth between instances)
  • Cons: if rack fails then all the EC2 instances will fail at the same time
  • Usage: Big data job that needs to complete fast, Application with low latency and high throughput

Spread

  • All EC2 instance will be located on different hardware
  • Span across multiple AZ
  • Reduced risk of simultaneous failure
  • Limit of 7 instances per AZ per placement group
  • Usage: Application that needs maximize HA , Critical Applications that needs to be isolated from failure from each other

Partition

  • Upto 7 partitions per AZ with 100s of EC2 instances
  • The instances of 1 partition do not share racks with instances of other partitions
  • A partition failure can affect many EC2 instances from same partition but it won't affect other EC2 instances on other partitions
  • EC2 instances can get access to the partition information using metadata
  • Usage: HDFS, HBase, Cassandra, Kafka