Failure types, write-ahead logging, deferred vs immediate update, checkpoints and how to decide which transactions to redo and which to undo after a crash.