Transactions on the Logical Database - Transaction Processing - page 23

Bibliographical Notes
The foundations of transaction-oriented database processing were laid by Eswaran
et al. [ 1974 , 1976 ] and Gray et al. [ 1976 ]. Transactions with partial rollbacks
were already implemented in the System R relational database management system
described by Astrahan et al. [ 1976 ], Gray et al. [ 1981 ], Blasgen et al. [ 1981 , 1999 ],
and Chamberlin et al. [ 1981 ]. The authors of System R state that the execution
of a transaction should be atomic, durable, and consistent, where the last property
is described as: “the transaction occurs as though it had executed on a system
which sequentially executes only one transaction at a time,” that is, in isolation
[Gray et al. 1981 ].
Gray [ 1980 ] presents a general formal model for transactions and their process-
ing. Gray [ 1981 ] reviews the transaction concept and implementation approaches,
analyzes the limitations of ordinary “flat transactions,” and argues for the need
of “nested transactions” as a means of modeling long-lived workflows. In their
textbook on the implementation of transaction processing systems, Gray and Reuter
[ 1993 ] also analyze thoroughly the transaction concept and review the history of the
development of the transaction concept and transaction processing systems.
In many textbooks on database management and transaction processing, most
notably in the classic works by Papadimitriou [ 1986 ]andBernsteinetal.[ 1987 ],
the read-write transaction model (Sect. 1.7 ) was adopted as the basis for discussing
transaction-oriented concepts. In this setting the database was considered to be a
set of uninterpreted abstract data items x that could be read (by action RŒx)and
written (by action WŒx). Schek et al. [ 1993 ] consider the enhanced model in which
the backward-rolling phase of an aborted transaction is represented explicitly as a
string of undo actions; this feature is also included in the read-write model (also
called the page model) used in the textbook by Weikum and Vossen [ 2002 ].
The key-range transaction model defined in this chapter and used previously by
Sippu and Soisalon-Soininen [ 2001 ], Jaluta [ 2002 ], and Jaluta et al. [ 2003 , 2005 ,
2006 ] was inspired by the model used by Mohan [ 1990a , 1996a ] and Mohan and
Levine [ 1992 ] to describe the ARIES/KVL and ARIES/IM algorithms. The transaction
model considered by Gray et al. [ 1976 ] included the feature that a transaction
can declare an update as committed before the transaction terminates, although
this feature was only allowed for transactions run at the lowest isolation level
(Problem 1.5 ).