Relational DBMS

In relational bases data are stored in a type of the tables consisting of lines and columns. Each table has own, predeterminated set of the referred to as fields. Columns of tables of relational base may contain scalar data of the fixed type, for example numbers, lines or dates. Tables in a relational database can be connected by the relations "one-to-one" or "one-to-many". The quantity of lines of entries in the table is unlimited, and each record corresponds to a separate entity.

Relational databases hold a dominant position now. Hierarchical and network structures of databases consigned to the past, having given way to relational bases under which postoroyeno the majority of modern DBMS (A MS SQL Server, MS Access, InterBase, FoxPro, PostgreSQL, Paradox and others).

Details

The relational model is focused on data structure in the form of two-dimensional tables. Each relational table represents a two-dimensional array and has the following properties:

• Each element of the table is one data member
• Each column has the unique name
• Identical lines in the table are absent
• All columns in the table homogeneous, i.e. all elements in a column have identical type
• The sequence of lines and columns can be any

The relational DBMS focused on implementation of the systems of operational data processing are less effective in problems of analytical processing, than multidimensional databases. It is connected, first, with existence enough tight restrictions imposed by the existing implementation of the SQL language. An example of such real-life restriction is the assumption that data in relational base are not arranged (or more precisely, are arranged in a random way). At the same time their ordering requires additional costs of time for sorting at each addressing the database. In the analytical systems input and data sampling is performed by large portions. In turn data after they are included in the database, remain invariable during the long time frame. And here data storage in the form of partially denormalized tables in which for performance improvement not only the detailed, but also previously calculated aggregated values can be kept appears more effective. And for navigation and selection methods of addressing and indexation can be used specialized, based on the assumption of small variability and low-mobility of data in the database. Such method of data structure, sometimes call precalculated, emphasizing with that, its difference from the normalized relational approach assuming dynamic calculation of a different type of results (aggregation) and establishment of communications between details from different tables (connective operation).

Main shortcomings

In addition to low efficiency about which it was told earlier it is possible to carry the fact that as the basic and, often, the only mechanism providing quick search and selection of separate lines to the table (or in the tables connected through foreign keys), different modifications of the indexes based on B-trees are usually used to shortcomings of traditional relational DBMS. Such solution is effective only when processing small groups of records and high intensity of data modification in databases.

Relational DBMS, perhaps, will never leave a scene, but days of their domination are definitely considered, Paul Kril who published in September, 2011 article about it in InfoWorld believes. He quotes the analyst Robin Blor who claims that the architecture of relational DBMS became morally outdated as it was created during last era and does not meet the modern requirements.

Relational DBMS still dominate in processing systems of financial transactions, but today the companies apply DBMS of new architecture NoSQL more and more widely — horizontally scalable, distributed and developed in open codes. Examples of such systems there is Hadoop, MapReduce and VoltDB. By estimates of analysts of Forrester, about 75% of data at the enterprises it or the semi-structured information (XML, e-mail and EDI), or unstructured (the text, images, audio and video), and only 5% of these data the rest is stored in relational DBMS, and — in bases of other types or in the form of files, and it is not subject to processing by the relational systems.

According to Blor, relational DBMS "can die so that nobody will notice it" — for example if Oracle in the DBMS simply replaces the SQL mechanism with NoSQL. Such mechanism, the analyst considers, there could be one of the column DBMS existing today.