Column generation-based algorithm for fragment allocation: minimizing query splitting in distributed databases

• Published in: Information technology and management Vol. 26; no. 4; pp. 559 - 572
• Main Author: Amiri, Ali
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2025; Springer Nature B.V
• Subjects: Algorithms; Business and Management; Cloud computing; Communication; Computer Communication Networks; Concurrency control; Costs; Data Structures and Information Theory; Design; Design optimization; Designers; Fault tolerance; Heuristic; Integer programming; IT in Business; Linear programming; Mathematical programming; Operations Research/Decision Theory; Queries; Response time (computers); Splitting; Traveling salesman problem; Integer programming; Column generation; Fragment allocation; Query splitting; Distributed databases
• ISSN: 1385-951X; 1573-7667
• DOI: 10.1007/s10799-024-00425-2

This paper focuses on the fragment allocation problem in distributed databases and proposes an approach that minimizes query splitting. Query splitting occurs when a query has to access multiple servers to retrieve the fragments it needs, resulting in reduced system performance. The objective of minimizing query splitting is important because it captures many factors that affect the performance of the database, such as reducing response time and cost. The paper presents a column generation-based algorithm to solve the fragment allocation problem, which requires less fine-tuning of its parameters and outperforms the IP approach implemented by CPLEX in terms of the number of queries split and execution time. The approach and algorithm offer practical solutions to optimize the design of a distributed database system. The paper’s contribution is significant as it fills the gap in the literature by offering a novel approach that minimizes query splitting, which can serve as a proxy for achieving a combination of other objectives such as minimizing costs, reducing response time, and balancing server workloads.