Edgar F. Codd

Edgar Frank "Ted" Codd (19 August 1923 — 18 April 2003) took the idea that how data is stored should be separable from how it is asked about, and raised it, largely alone, to the level of mathematics. The 1970 CACM paper was the starting point, not the achievement. Most of the rest of his working life went into persuading the industry — beginning with his own employer — to accept it.

Background

Codd was the youngest of seven children, born on the Isle of Portland in Dorset, on the south coast of England. His father manufactured leather; his mother taught school. He went up to Exeter College, Oxford, on a full scholarship to read chemistry, but in 1942, though eligible for deferment as a student, he volunteered and served as a flight lieutenant in RAF Coastal Command, flying Sunderlands. He returned after the war, switched to mathematics, and took his degree in 1948.

He emigrated to the United States that year, worked briefly at Macy's in New York, taught mathematics at the University of Tennessee for six months, and joined IBM in 1949 as a programming mathematician on the Selective Sequence Electronic Calculator. He went on to work on the design of the IBM 701. In 1953 he left both IBM and the country in protest at Senator Joseph McCarthy, running the data processing department of Computing Devices of Canada in Ottawa. A chance meeting with a former manager brought him back to IBM in 1957, at Poughkeepsie, where he worked on STRETCH (the IBM 7030) and led the team that built what is generally described as the world's first multiprogramming system. From 1961 he studied at the University of Michigan on an IBM scholarship, completing his doctorate in 1965; his thesis simplified von Neumann's self-reproducing automata, reducing the required 29 states to eight, and was published in 1968 as Cellular Automata.

1968 to 1972

Codd transferred to IBM's San Jose Research Laboratory in 1968 and turned to databases. An internal report, Derivability, Redundancy and Consistency of Relations Stored in Large Data Banks, appeared in 1969; the CACM paper followed in 1970.

The 1970 paper is not the whole of his theory, and it is regularly credited with more than it contains. It defines five operations on relations — permutation, projection, join, composition, restriction — and, on the language question, goes only as far as proposing a universal data sublanguage based on an applied predicate calculus. The phrases "relational algebra", "relational calculus" and "functional dependency" appear nowhere in it. Normalization recast in terms of functional dependencies is Further Normalization of the Data Base Relational Model (1971); the equivalence of the algebra and the calculus is Relational Completeness of Data Base Sublanguages (1972). What textbooks call Codd's theory accumulated over three years.

Fifteen years inside IBM

IBM was cool to it. The biography attached to his Turing Award record, written by C. J. Date, gives the reason plainly: the company was heavily invested in its existing non-relational product IMS and anxious to keep that revenue, and was at first unreceptive to the point of hostility. The consequence was that Relational Software Inc. (later Oracle Corporation) and Relational Technology Inc. (later Ingres Corporation) brought relational products to market well before IBM did — Oracle V2 shipped in June 1979, while IBM's SQL/DS came in 1981 and DB2 in 1983.

Nor did those IBM products satisfy him. System R adopted SEQUEL, later SQL, rather than Alpha, the query language Codd had designed; in his judgement IBM's support for the relational model was incomplete and in places simply wrong. He resigned from IBM in 1984.

The rules

On 14 and 21 October 1985 Codd published a two-part article in Computerworld. The first was headed "Is your DBMS really relational?" It was a counter-attack on the sudden abundance of vendors calling their products relational, and it supplied a test. The set is universally called the twelve rules, but there are thirteen of them, numbered from zero. Rule Zero demands that a system claiming to be a relational database management system "must be able to manage data bases entirely through its relational capabilities". The second instalment applied the test to three shipping products.

In his later work he extended the model toward analysis, coining the term OLAP (On-Line Analytical Processing).

Awards and honours

• 1974 Fellow of the British Computer Society
• 1976 IBM Fellow
• 1981 ACM Turing Award, "for his fundamental and continuing contributions to the theory and practice of database management systems"
• 1981 Elected to the National Academy of Engineering
• 1996 IEEE Computer Pioneer Award
• In 2004 ACM SIGMOD renamed its annual innovations award the SIGMOD Edgar F. Codd Innovations Award. SIGMOD's predecessor body, SICFIDET, had been founded by Codd himself in 1970

Legacy

Codd died on 18 April 2003 at Williams Island, Florida. The relational database industry he had made possible grew into tens of billions of dollars a year; he never benefited directly from it.

What he left was not a product but a criterion. Before Codd, "is this a good database system?" was a matter of opinion; after him, "is this really relational?" was a question with a test attached. That is why database engineering became a scientific discipline rather than a collection of techniques. SQL was not his design, so his victory was never complete. But cash machines and airline reservations still run inside the frame he drew in 1970.