JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY HYDERABAD B.Tech III DISTRIBUTED DATABASES Previous Question Paper (March - 2021 - R18)

Fragmentation is the process of partitioning a global relation into smaller fragments to improve performance and data locality. There are two primary strategies:

Horizontal Fragmentation: This strategy partitions a relation by selecting specific rows (tuples) based on a predicate. Each fragment contains a subset of the tuples of the original relation, but all attributes are preserved. For example, in a global table 'Employee' with attributes (EmpID, Name, Dept), a horizontal fragment for the 'Sales' department would include all attributes for only those rows where Dept = 'Sales'.

Vertical Fragmentation: This strategy partitions a relation by selecting specific columns (attributes) based on their usage patterns. Each fragment contains a subset of attributes, but all tuples are preserved. A primary key must be included in every fragment to allow for reconstruction via natural join. For example, in the same 'Employee' table, one fragment might contain (EmpID, Name) for personal records, while another contains (EmpID, Salary) for payroll processing.

The reference architecture for distributed databases (often based on the ANSI/SPARC model) provides a framework for managing data distribution and independence through several key features:

• Global Conceptual Schema (GCS): Provides a unified view of the entire database, hiding the fact that data is distributed across multiple sites.
• Fragmentation Schema: Defines how relations are partitioned into fragments. It maps the GCS to the fragments, ensuring that the user does not need to know where specific data resides.
• Allocation Schema: Maps the fragments to specific physical sites in the network. This layer manages data replication and placement to optimize query performance.
• Local Mapping Schema: Defines the mapping between the fragments and the local database management systems (LDBMS) at each site, allowing for heterogeneity in local storage.
• Data Independence: The architecture ensures both logical and physical data independence, meaning changes to the distribution or local storage do not affect the global user queries.