“The Transaction Concept: Virtues and Limitations” by Jim Gray was an invited paper at the Very Large Data Bases (VLDB) conference in 1981, and was also issued as Tandem technical report TR 81.3. It is the paper most often credited with turning the transaction from an implementation detail into a clearly defined programming abstraction that application writers could rely on.
The paper defines a transaction as a transformation of state that has the properties of atomicity, durability, and consistency: atomicity means the changes are all or nothing, durability means a committed transaction’s effects survive failures, and consistency means each transaction is a correct transformation of the database state. These properties are the ones later assembled, with isolation, into the ACID acronym.
Beyond the definition, Gray surveys how transactions are implemented and where the concept can be extended, discussing techniques such as logging and locking and considering nested and long-lived transactions. The “limitations” in the title reflect his honest accounting of where the simple model strains, foreshadowing decades of work on distributed and long-running transactions.
The paper drew on Gray’s experience building real systems, including IBM’s System R, and it became a touchstone for database and distributed-systems research. Its framing of the transaction as a contract between application and system remains the way the concept is taught and used today.