Schema Transformation from RDBMS to NoSQL Using Hypergraph

Schema Transformation from RDBMS to NoSQL Using Hypergraph

연구 분야: Databases

학회: International Conference on Big Data Analytics

The current study explores methodologies (like TLD, CLDA, and Kuzera) and the challenges of migrating schemas from RDBMS to NoSQL document stores. These challenges include data misinterpretations, increased response time due to costly joins, poor data locality, and unalignment of schema design with query patterns. To encounter these challenges, the proposed model presents a schema transformation approach from a Relational Database Management System (RDBMS) to NoSQL using Hypergraph. The existing methodologies use a graph data model that cannot represent complex relationships, whereas the Hypergraph provides accurate group-wise complex relationships. The paper contains four phases: input, data modeling decision, transformation, and output phase. The input phase contains relational tables and workload queries. The data modeling decision phase incorporates the calculation of normalized weight followed by the decision to select the type of data model (embedding or referencing). The role of the transformation phase is to generate a collection of tables that reflect the table hierarchy, utilized to create the NoSQL schema. The transformation phase comprises hypergraph generation, matrix transformation, and matrix enumeration. The hypergraph generation phase includes query mapping, inverse query mapping, and hypergraph generation. The matrix transformation phase covers the creation of the incidence matrix and square matrix, which are used for matrix enumeration. The incidence matrix shows which queries are related to which tables. The square matrix represents the frequency and strength of these relationships. Matrix enumeration is used to finalize the grouping of related tables based on query patterns, optimizing the schema for better performance by reducing the need for complex joins. In the output phase, the final NoSQL schema is created by using the results of the preceding phases (data modeling decision phase and transformation phase). An experiment has been conducted on the TPC-H benchmark and NoSQL document-oriented datasets. The proposed model is compared with existing models (TLD, CLDA, and Kuszera), where it outperforms TLD with 6.67% and Kuszera with 12.50% in terms of time efficiency. However, CLDA worked better than the proposed model due to additional computational overhead for managing the hypergraph structure.