A theoretical formulation for degrees of isolation in databases

Abstract

Although isolation is one of the desirable properties, most commercial database management systems do not provide complete isolation to transactions. They offer different degrees — 0, 1, 2, or 3 — of isolation, to transactions. By providing lower degrees of isolation, response time of a database system can be improved, although at the expense of consistency. Originally different degrees of isolation were defined in terms of lock-based protocols. This paper formulates these different degrees of isolation in terms of histories, as in the case of the usual serialization theory and proposes timestamp-based protocols for different degrees of isolation.