Database consistency under UNIX

Abstract

A recovery technique to facilitate the recovery from system failures should ensure a consistent database state at all times. The UNIX write system call uses a delayed-write policy. Data blocks are kept in a buffer cache and are later asynchronously written to disk. If a system failure occurs, logically complete transactions may be lost and the database may contain incorrect old data values. If a physical write was being executed when the failure occurred, the database will be inconsistent, since an update in place policy is used when writing to disk. Differential files can provide database consistency efficiently.