• Atomicity: All ops happen, or none
• Consistency: Data will be consistent, no constraints violated
• Isolation: One transaction cannot read from another non-completed one. If t2 reads write from t1, needs to wait.
• Durability: Changes have been recorded

“In presence of a network partition, one has to choose between consistency and availability”.

• Strong consistency
• Timeline consistency: Typically async master-slave replication.
• Eventual consistency: Temporarily inconsistent

Scaling out NOSQL across regions/public clouds. Partitioned db across clusters. How to ensure that transaction is commited on all nodes.

Ensuring atomicity: Multi-phase commit, pessimistic read-write locking But: Network partitions