As mentioned in Section 1, ARIES uses Write-Ahead-Logging, a Steal/No-Force policy in the Buffer Manager and repeats history during recovery. Each page in the database contains a log sequence number (LSN), which uniquely identifies the log record for the latest update to the page. This page LSN determines whether or not recovery is to be done to that page. This LSN information is used to determine the point in the log from which the Redo phase must commence during the restart from a system crash. LSNs are often implemented using the physical address of the log record in the log.
During the Analysis phase, the log is processed from its most recent checkpoint. This phase determines information about the dirty pages and active transactions that is used in the subsequent phases. The second phase is the Redo phase, in which history is repeated by processing the log from the earliest log record that could require a redo, ensuring that all logged operations have been applied. The third phase is the Undo phase. This proceeds backwards from the end of the log, removing from the database the effects of all transactions that had not been committed at the time of the crash. During the Undo pass, Compensation Log Records (CLR) are written to ensure ARIES need never undo the effects of an Undo operation.