Database concurrency control theory characterizes the situations in which the concurrent execution of transactions is acceptable. But concurrency control theory does not deal with transaction abort. The theory developed in this paper characterizes the additional restrictions necessary to ensure that transaction abort does not change the effects of committed transactions. It also characterizes the information that must be written to a log in order to provide recovery from transient system failures.
The paper is quite formal, and has a high ratio of definitions to theorems. The paper’s value to practitioners is in providing a framework for and examples of reasoning about database update and recovery techniques.