Overview of Distributed Database Systems: Concepts, Classification, Features, and Data Sharding

With the maturation of traditional database technology and the rapid development of computer networking, database applications are now commonly built on networked environments.

Centralized database systems exhibit shortcomings such as high communication overhead, low reliability due to single‑point failures, and limited scalability, prompting a shift toward distributed computing models, especially client/server and distributed database architectures.

Distributed database systems, a fusion of database and network technologies, emerged in the mid‑1970s; the first system, SDD‑1, was implemented in 1979. Since the 1990s, they have become commercial products, evolving toward client/server models.

Distributed databases are classified into three types: homogeneous‑homogeneous (same data model and DBMS), homogeneous‑heterogeneous (same model, different DBMS vendors), and heterogeneous (different models and types across sites).

Key characteristics of distributed databases include physical distribution across multiple sites, logical unity perceived by global users, site autonomy, and collaborative operation among sites. Additional traits are data independence, combined centralized‑autonomous control, controlled redundancy, and distributed transaction management.

Advantages encompass flexible architecture, suitability for distributed management, superior economic performance, high reliability and availability, fast response for local applications, and good scalability and integration. Disadvantages involve significant communication overhead, complex access structures that may not be effective in distributed settings, and challenges in ensuring data security and confidentiality.

Data sharding techniques are categorized as horizontal (partitioning rows), vertical (partitioning columns), derived (based on attributes of other relations), and hybrid (combining methods). Sharding must satisfy completeness, reconstructability, and non‑overlap conditions.

Data allocation strategies include centralized (all fragments at one site), partitioned (each fragment placed at a specific site), full replication (complete copies at every site), and hybrid (between partitioned and replicated).

The architecture separates data sharding from data allocation, forming a “data distribution independent” concept, enables explicit redundancy control, and supports local DBMS independence (local mapping transparency).

A distributed database management system (DDBMS) handles user requests, determines target sites, accesses network data dictionaries, performs distributed processing when data reside on multiple machines, provides communication interfaces, and supports data and process migration in heterogeneous environments.

The concluding summary reiterates that distributed computing breaks the limits of centralized DBMS, that client/server structures have evolved toward multi‑tier designs, and that distributed databases now dominate the field, offering logical unity, flexible allocation, and improved performance while introducing new challenges in concurrency control, recovery, and network efficiency.