- One cassandra instance per data disk per server
- One disk per zfs pool
- SSD shared between the cassandra instances
- All the cassandra instances of a server must be consider in the same rack to avoid to have all the replicas in the same server
- 5 nodes at the beginning of the tests
- add the last node during the test
Additional test scenarios:
- Loss of a cassandra instance in a server
- Loss of several instances in different servers
- Loss of a complete server