On Five De nitions of Data Integrity 1 Ravi

This paper compares ve deenitions of data integrity, and shows how they can be ordered in an increasingly restrictive sequence. The most general of these, due to Courtney and Ware 6], is based on the concept of expectation of data quality: data has integrity to the extent that its quality meets, or exceeds, the quality requirements that users expect of it. This deenition incorporates liveness requirements, whereas the others only address safety requirements. The second and third deenitions are both based on the ability to modify data. One of these, due to Sandhu and Jajodia 16], deenes the scope of integrity to be safeguards against the improper modiication of data. The other narrows the scope further to safeguards against the unauthorized modiication of data. This latter deenition has been popular in recent security criteria 4,10]. The fourth deenition discussed here is Biba's concept of integrity as one-directional information ow in a lattice 2]. We argue that Biba's deenition is more restrictive than the previous three, being the one in which unauthorized modiication is given a very speciic meaning. The fth and most restrictive deenition comes from the network arena and requires that the data not be modiied (or at least that any change should be detectable). The paper concludes with a discussion of how these deenitions relate to other work in the integrity arena.