Consequences of the problem

A using program is not interested in the bits stored in a file for themselves, but generally uses them to represent some useful logical structure. For example, an accounting program may intend to update a number of accounts to reflect a sequence of transactions. The data in one or more files may represent the state of the accounts, how much money each has, while another file may represent the sequence of transactions, ``move so many dollars from one account to another''. If this accounting program took no precautions, and a crash occurred while the accounts were being updated, the contents of the represented accounts after recovery will have no predictable relationship to the contents before the crash. A simple solution would be to have backups for all the files, and return to the backups after a crash. There still remains the question of how to identify which files are current. If this information is maintained in files (where else?), and a crash occurs while it is updated, chaos may still occur. If the accounting records are extensive, and the transaction file is long, it may be too expensive to maintain a complete backup of all accounting records while the transaction file is processed.