[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

Journal of Information Science and Engineering, Vol. 34 No. 1, pp. 243-259

Migration from RDBMS to NoSQL Using Column-level Denormalization and Atomic Aggregates

1Positioning/Navigation Technology Research Section
Intelligent Cognitive Technology Research Division
Electronics and Telecommunications Research Institute
Daejeon, 34129 Korea
E-mail: jjryu@etri.re.kr

2Department of Computer Engineering
Kwangwoon University
Seoul, 01897 Korea
E-mail: {kihoonlee, pointnb}@kw.ac.kr

As NoSQL grows in popularity, many organizations are attempting to migrate their databases from RDBMS to NoSQL. Because NoSQL is very different from RDBMS, such migration is a challenging issue. For example, NoSQL does not support join operations or transactions. We propose a novel solution for database migration from RDBMS to document-oriented NoSQL, which is the most widely used type of NoSQL. Our method not only avoids join operations with a marginal increase in the database size, but also supports atomicity using the notions of column-level denormalization and atomic aggregates. Column-level denormalization duplicates only columns that are accessed in non-primary-foreign-key-join predicates. Atomic aggregates combine tables that are modified within the same transaction into a unit of atomic updates called an aggregate. Experimental results using TPC-H and MongoDB show that our method improves the query performance by up to 2.2 times using 1.5 times more space compared with a baseline method that uses the relational schema as it is. Compared with the state-of-the-art method, which duplicates whole tables to avoid join operations, our method improves the query performance by up to 2.0 times with 2.8 times less space.

Keywords: database migration, column-level denormalization, atomicity, RDBMS, No-SQL

  Retrieve PDF document (JISE_201801_15.pdf)