第8章データ管理

Data models can be conceptual (represented using EER or UML notations), logical, physical, or self-describing. The various logical data models can be categorized as follows:

• A hierarchical DBMS is a data model in which the data are organized into a treelike model, DML is procedural and record-oriented, the query processor is logical, and internal data models are intertwined (e.g., IMS from IBM).

• A network DBMS is a data model in which the data are organized into a network model, DML is procedural and record-oriented, the query processor is logical, and internal data models are intertwined (e.g., CA-IDMS from Computer Associates).

• A relational DBMS is a data model in which the data are organized into a relational data model, use SQL as a declarative and set-oriented database, the query processor has a strict separation between the logical and internal data model. Relational DBMSs are the most popular in the industry (e.g., MySQL open-source database from Oracle, Oracle DBMS, DB2 from IBM, and Microsoft SQL).

• An object-oriented DBMS is a data model in which the data are organized into an OO data model, avoiding impedance mismatch when used with an OO host language. It is also called OODBMS or ODB (e.g., db4o open-source database from Versant, Caché from Intersystems, GemStone/S from GemTalk Systems, which are only successful in niche markets, due to their complexity).

• An XML DBMS is a data model in which the data are using the XML data model to store data. XML could be native XML DBMS (e.g., BaseX and eXist), which map the tree structure of an XML document to a physical storage structure, or XML-enabled DBMS (e.g., Oracle and IBM Db2) are existing DBMS that are extended with facilities to store XML data.

• A not-only SQL DBMS, or NoSQL DBMS, comes in a variety of big unstructured data classified as a document, a graph, key-value stores, and column-oriented databases. NoSQL DBMS focuses on scalability and the ability to cope with irregular or highly volatile data structures (e.g., Apache Hadoop, MongoDB, Neo4j). The irregular data structures appear when the data don’t follow a specific order or nature. The volatile data are usualy stored in cache memory and is easy to lose.

With a single-user DBMS, only one user at a time can use the database. A multiuser DBMS allows many users to use the database concurrently. A multiuser DBMS is illustrated in Figure 8.8.

DBMS users may be divided into actors on the scene who use and control the database content; design, develop, and maintain database applications such as database administrators and database designers; and workers behind the scenes who design and develop the DBMS software and related tools as well as the computer systems operators, including system designers and developers.

There are various types of database architectures as follows: centralized DBMS architecture, n-tier DBMS architecture, cloud DBMS architecture, federated DBMS, and in-memory DBMS. In centralized DBMS architecture, the data are maintained on a centralized server at a single location while client-server DBMS architecture has one or more client computers connected to a central server over a network. Active clients request services from passive servers. An n-tier DBMS architecture usually divides an application into three tiers/layers: the presentation tier, the logic tier, and the data tier (e.g., a client with GUI functionality, an application server with applications, a database server with a DBMS and a database, and a web server for web-based access). In cloud DBMS architecture, the DBMS and database are hosted by a third-party cloud provider (e.g., Apache Cassandra project and Google’s BigTable). A federated DBMS provides a uniform interface to multiple underlying data sources which hide the underlying storage details to facilitate data access. An in-memory DBMS stores all data in internal memory instead of slower external storage (e.g., disk; often used for real-time purposes such as HANA from SAP).