Rethinking integrity [distributed databases]

In practice, no real database enjoys complete integrity, a variety of factors conspire to make some information stale, missing, or just plain wrong. System design and analysis should recognize that real systems lack integrity to some degree, most, if not all, of the time. Significant benefits flow from such recognition. Risk management techniques can identify the severity of different integrity-loss scenarios, thereby focusing scarce resources on critical areas. A designer can deliberately sacrifice nonessential integrity under carefully controlled conditions to achieve other design objectives, such as performance, autonomy, availability, or security. Designers can achieve these objectives and still preserve essential aspects of integrity. The authors discuss the factors that undermine integrity in distributed databases