What is Checkpoint?

The original idea comes from the ARIES paper (see reference). For performance reason, database does not persist modified data to disk after every committed change, instead, it periodically issues checkpoint which persists the changes in batch. It's called checkpoint because when database crashes and then starts again for crash recovery, the recovery process only needs to recover the committed changes after the latest checkpoint because the checkpoint guarantees any commits before that checkpoint have been persisted already. Checkpoint is still a heavy operation and may impact database performance.