MASTERMIND
EDITOR: George Strawn, gostrawn@[Link]
Relational
Databases: Codd,
Stonebraker, and
Ellison
George Strawn and Candace Strawn
T
he desirability of creat- Codd, Michael Stonebraker, and that row. A data table is also called
ing high-level languages Larry Ellison. Finally, we discuss a relation because, mathematically
to ease the programming new developments in databases speaking, a table is a subset of the
task was recognized soon that point to a future “beyond re- Cartesian product of its attribute
after the creation of comput- lational.” sets, and that’s the definition of a
ers, and by the late 1950s, Algol, relation.
Fortran, and Cobol were in use. Relational Database Relational DBMSs are used to
High-level data languages were Systems store and process the primary
slower to emerge, perhaps be- Early approaches to referencing information of an organization,
cause early computers were quite data in programming languages including transactions, ad hoc
limited in the amount of data they did not separate the logical de- queries, and reports. An RDBMS
could store and manipulate. In the sign of the data from its physi- is usually made up of many tables,
1960s, the computing community cal representation. On the other which can be related to each other
began defining high-level data hand, the relational model pro- by including keys from one table
languages through activities such vided a clear separation: data are in another. The included keys are
as the Conference/Committee on designed as tables (superficially called foreign keys. To continue the
Data Systems Languages.1 Then, like today’s spreadsheets), and earlier example, if there is a de-
in 1970, an article appeared in the it is up to the database manage- partment table whose attributes
Communications of the ACM titled ment system (DBMS) to deter- include department number and
“A Relational Model of Data for mine how to store those tables in department name, the department
Large Shared Data Banks,”2 by the computer’s memory. A data number in the employee table is a
Edgar F. Codd. This article would table has rows that represent dis- foreign key. An important (and
determine the primary direction tinct entities (such as employees) difficult to implement efficiently)
of high-level database languages and columns that represent various operation is the join, which could
for the next 40 years. Here, we attributes of those entities (such produce a composite table whose
first give a brief overview (not a as employee number, name, sal- rows combine the employee infor-
tutorial) of the relational model ary, and department number). mation with the appropriate de-
for database systems, then high- Each row must have a key attri- partment information.
light three of the masterminds bute (employee number, in this Other important operations,
who created and developed it: example) that uniquely identifies such as insert, delete, and update (a
1520-9202/16/$33.00 © 2016 IEEE Published by the IEEE Computer Society [Link]/ITPro 63
MASTERMIND
row in a table), were added in the create a relational database proto- [Link]/wiki/Michael_
mid-1970s, when SQL was added type ([Link] Stonebraker). Upon graduation,
to the relational model. These wiki/IBM_System_R), Codd was he became a faculty member at the
operations enable an RDBMS to excluded from the development University of California, Berkeley,
do transaction processing—for team. The prototype was named where he remained for 29 years. In
example, keeping a company’s in- System R, and it was here that the 1973, after reading Codd’s papers
ventory or sales up to date. This SQL language was created. Codd on the relational model, he and a
is a complex process that wasn’t was disappointed because, in his colleague decided to attempt to
fully understood until the early opinion, it wasn’t “fully relational.” build a practical relational database
1980s, when the necessary char- According to the Wikipedia ar- system. Because of the intentional
acteristics of a transaction were ticle on Codd, separation of the relational model’s
defined (atomicity, consistency, design from implementation con-
isolation, and durability). Another As the relational model started to siderations, there was plenty of
important part of the relational become fashionable in the early work to do to prove that a practical
model defined normal forms of 1980s, Codd fought a sometimes implementation was even possible.
tables that minimized storage re- bitter campaign to prevent the Stonebraker’s work proceeded in
quirements and simplified update term being misused by database parallel with IBM’s development of
operations. In the late 1980s, the vendors who had merely added a System R. By the mid-1970s, he and
relational model was extended to relational veneer to older technol- a group of student programmers
distributed databases, although, ogy. As part of this campaign, he had produced a working RDBMS,
as discussed later, this extension published his 12 rules to define which they named Ingres, for in-
would expose a limitation of the what constituted a relational da- teractive graphics and retrieval system
relational model. tabase [[Link] ([Link]
Ingres_(database)). Ingres was imple-
mented on minicomputers, whereas
System R was a mainframe develop-
Because of the intentional separation of the ment. By the early 1980s, the gap
relational model’s design from implementation between mainframe and minis was
narrowing, and Ingres was viewed
considerations, there was work to do to prove that as a serious RDBMS prototype “in
a practical implementation was possible. competition with” System R. At
this point, Berkeley’s open source
software licensing policy became
The Creator: wiki/Codd%27s_12_rules]. This important ([Link]
Edgar F. (Ted) Codd made his position in IBM in- org/wiki/BSD_license).
Codd (1923–2003) was an Eng- creasingly difficult, so he left to In the late 1970s, Berkeley com-
lishman who studied mathematics form his own consulting company puter scientists had developed a
and chemistry at Oxford and then with Chris Date and others. (See version of the UNIX operating
served as an RAF pilot in World [Link] system that they named Berkeley
War II. In 1948, he immigrated wiki/Edgar_F._Codd.) Standard Distribution (BSD). This
to the US and worked for IBM as system was made widely avail-
a mathematical programmer. In The computer science research able under a liberal and low-cost
1965, he earned a PhD in computer community was an enthusiastic license. That license was subse-
science at the University of Michi- supporter of the relational model, quently used by other software
gan. In 1967, he moved to IBM’s and Codd was the recipient of the development projects at Berkeley,
San Jose Research Laboratory in 1981 Turing Award. including the Ingres RDBMS.
California. Three years later, he This code became the basis for a
published the relational model for The Developer: number of commercial develop-
databases. IBM was slow to rec- Michael Stonebraker ments, including one launched by
ognize the commercial potential Stonebraker (1943–) earned his Stonebraker and colleagues. This
of the relational model, perhaps bachelor’s degree at Princeton process—creating and then com-
in part because it had other data- in 1965 and, like Codd, received mercializing a number of database
base products that it didn’t want to his graduate education at the systems—has characterized Stone-
disrupt. When it finally decided to University of Michigan (https:// braker’s long career, first at Berkeley
64 IT Pro March/April 2016
and then at MIT after 2001. His ef- By the beginning of the 21st cen- of NoSQL (not SQL) models
forts have been recognized by the tury, Oracle had achieved a secure ([Link]
computer science research com- place for itself in the RDMBS world, NoSQL). One relatively well-known
munity, and he was the recipient of along with IBM’s DB2 and Micro- example is the database model that
the 2015 Turing award. soft’s SQL Server. It then expanded enables Google to process the entire
its scope of operation by acquiring Web (continuously!) to give us the
The Entrepreneur: Sun Microsystems in 2010. Ellison, ability to effortlessly search it. The
Larry Ellison as an entrepreneur, has never re- software to perform this giant task
Both Codd and Stonebraker fol- ceived a Turing award, but he has is known variously as MapReduce
lowed the research path to da- been honored by the stock market. or Hadoop ([Link]
tabase fame. Ellison (1944–) fol- His RDBMS-created net worth is org/wiki/Apache_Hadoop). It would
lowed the Bill Gates and Steve Jobs about US$50 billion, which should be hopeless to try to use an RDBMS
path of dropping out of college be honor enough! for this task.
([Link]
Larry_Ellison). Also like Jobs, Beyond Relational
D
Ellison grew up with adoptive par- RDBMSs have dominated the da- atabase models with
ents. While attending the Univer- tabase world for decades, but they names like column, doc-
sity of Chicago, he took a computer have never been the exclusive mod- ument, key-value, graph,
course, which might have deter- el for database development. Before and multi-model have emerged.
mined his post-dropout path. He relational, there were the hierar- Chances are that some of these
moved to California in 1966 and chical and networked models. The models (and probably others) will
worked briefly for Amdahl before latter have faded away, but hierar- serve the big data needs of the fu-
joining Ampex. While at Ampex, chical models have experienced re- ture. And because big data under-
he got involved with database proj- newed life with XML databases. A lies many other important areas,
ects and, like Stonebraker, was in- newer type, called object-oriented, such as the Internet of Things,
spired by Codd’s seminal paper. In arose in the 1980s and still com- machine learning, and artificial
1977, he founded a company to de- mands attention because of its intelligence, the database devel-
velop relational databases, which, compatibility with object-oriented opments of the future might be
after two name changes, was programming languages. even more important than those
called the Oracle Corporation. The biggest push for new data- of the past.
Ellison knew that IBM was de- base models has come from hard-
veloping System R, and he hoped ware developments. Disk storage References
his system would achieve com- has experienced a Moore’s law 1. T.W. Olle, The Codasyl Approach to
patibility with it. Not surprisingly, level of improvement over the Data Base Management, Wiley, 1978.
IBM declined the opportunity to years. For example, in 2015, a 2. E.F. Codd, “A Relational Model of
enable that compatibility. Ellison multi-terabyte disk for a PC cost Data for Large Shared Data Banks,”
pressed ahead anyway, releas- less than $100. Connecting thou- Comm. ACM, vol. 13, no. 6, 1970, pp.
ing the first version of the Oracle sands or more of these disks cre- 377–387; doi:10.1145/362384.362685.
RDBMS in 1979. As with most ates the potential for databases
startups, Oracle was initially of petabytes and even exabytes. George Strawn is the former director of
dogged by financial problems and Because disks can be connected the National Coordination Office for the
competitive pressures. In the ear- over the Internet as well as locally, Networking and Information Technol-
ly 1990s, Sybase, which was based distributed and federated databas- ogy Research and Development Program
on Stonebraker’s Ingres system, es are common these days. All of (NITRD). He is now retired. Contact
looked to be the RDBMS vendor these developments have contrib- him at gostrawn@[Link].
of choice until a corporate merger uted to the rise of the phrase (and
derailed it. The Sybase database the reality of) big data. Candace Strawn is a retired high school,
software was subsequently sold to It appears that RDBMS opera- community college, and university teacher.
Microsoft, who markets it as SQL tions might have a size limit (http:// Contact her at castrawn@[Link].
Server. The next competitor to [Link]/1PrhmR0). New database
battle Oracle for dominance was models that don’t provide all of the
Informix, but business missteps operations of the relational model Selected CS articles and
caused it to falter and be absorbed but do scale to big data sizes have columns are available for free at
by IBM. arisen under the group designation [Link]
[Link]/ITPro 65