JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY HYDERABAD B.Tech IV Year DISTRIBUTED DATABASES Previous Question Paper (February/March - 2022 - R18)

Architectural alternatives in a distributed database management system (DDBMS) are primarily categorized based on the distribution of data, the autonomy of local sites, and the degree of heterogeneity. The three main architectural models are:

1. Client-Server Architecture: This model separates the system into a centralized server that manages the database and multiple clients that request data. It is simple but can suffer from bottlenecks at the server.
2. Collaborative Server Architecture: This model allows multiple database servers to work together to execute a query. Each server manages its own local data and communicates with others to process distributed requests, providing better scalability.
3. Middleware Architecture: A middleware layer sits between the client and multiple heterogeneous database servers. It provides a unified global view of the data, translating queries into the specific language required by each underlying local database system.

Additionally, architectures are classified by autonomy (how much control local sites retain), distribution (where data resides), and heterogeneity (whether the systems use different software or hardware).

Partitioning (or fragmentation) is the process of dividing a logical database relation into smaller, non-overlapping subsets called fragments, which are then stored at different sites. The primary goal is to improve performance by placing data closer to where it is most frequently accessed. The main algorithms/types are:

1. Horizontal Partitioning: Rows of a relation are grouped based on a selection predicate. For example, a table of employees might be split by department ID, where all rows for 'IT' go to one site and 'HR' to another.
2. Vertical Partitioning: Columns of a relation are grouped. Each fragment contains the primary key to allow for reconstruction. This is useful when different applications require different subsets of attributes.
3. Hybrid Partitioning: A combination of horizontal and vertical partitioning, where a relation is first split horizontally and then those fragments are split vertically, or vice versa.

The algorithm must ensure completeness (all data is mapped to at least one fragment), reconstruction (the original relation can be recreated using relational algebra operators like Union or Join), and disjointness (fragments should not overlap, except for the primary key in vertical partitioning).