Enterprise Data Management for Transaction and Analytical Processing - Benchmarking Transaction and Analytical Processing Systems - page 15

The isolation property means that events happening during the execution of
a transaction are hidden from other transactions that are running concurrently.
Gray et al. [ 76 ] originally introduced isolation in four degrees of consistency :
(1) protecting other transactions from updates of a transaction, (2) additional
protection from losing updates, (3) additional protection from reading incorrect data
items, and (4) additional protection from reading incorrect relationships among data
items. The other three properties were later defined in [ 73 ]. Atomicity ensures that
all operations within a transaction are completed or none is. Consistency means
that the data is left in a consistent state after the normal end of a transaction has
been reached. Durability ensures that the results of a transaction upon successful
completion are preserved by the system and that they survive any subsequent
malfunction.
Early transaction processing systems already provided the functionality to
interactively execute very short transactions, in a so-called on-line processing mode.
This is where the term online transaction processing (OLTP) stems from. In online
transaction processing, the changes and results are immediately visible as opposed
to batch processing. In the beginning of the 1980s, the first commercial relational
databases that enabled interactive processing appeared. Similar to OLTP, the term
online analytical processing (OLAP) expresses that decision makers interactively
work with the system.
This chapter presents the foundations that this thesis builds on. It describes
the most relevant and widely used database design alternatives within the areas
of transactional and analytical processing. The understanding of database design
decisions taken in the past and the reasons behind facilitate building tomorrow's
hybrid systems and the benchmarks to evaluate and compare them.
Section 2.1 starts with the description of the data models used within
the databases for enterprise data management, explaining their characteristics.
Figure 2.1 presents an overview of past and today's most commonly used data
models and underlying database designs in OLTP and OLAP systems. Early data
management was handled via flat files. To avoid implementing the access and data
modification logic in each application, database management systems emerged
as an abstraction layer that provides standard interfaces to manipulate data. A
database management system (DBMS) is a collection of interrelated data and a
set of programs to access that data. The term database refers to the collection of
interrelated data, which contains information relevant to, e.g., an enterprise [ 191 ].
In OLAP systems dimensional modeling appeared along with the multidimensional
and hybrid data models to better match the users needs. Concerning the relational
data model, different database designs were introduced that are used in OLTP,
OLAP, or in both kinds of systems.
The relational data model emerged as today's most prevalent data model used in
business data processing in the areas of transaction processing as well as analytical
processing. Therefore, the scope of the subsequent sections is narrowed down to the
relational data model and the diversity of database designs within that area.
Section 2.2 covers database design alternatives and optimization opportunities in
the area of relational databases. For later discussions, it includes relevant aspects