Which Category Is Better: Benchmarking the RDBMSs and GDBMSs

Abstract

Relational database management systems (RDBMSs) have been a common option to manage structured data over the past decades. In recent years, with the prevalence of big data applications, vast unstructured and semi-structured data are generated, deeply challenging the relational model used in RDBMSs. For this reason, a wide spectrum of NoSQL databases are developed for managing unstructured, semi-structured or structured data. For example, graph database management systems (GDBMSs) are commonly used as an important category of NoSQL databases, to manage sophisticated graph data as well as relational data. Nonetheless, as claimed in existing literatures, both RDBMSs and GDBMSs are capable of managing graph data and relational data, the boundaries of them still remain unclear. In this paper, we propose a unified benchmark for RDBMSs and GDBMSs, to evaluate them under the same metrics, and report which category is better in different application scenarios. We conduct extensive experiments over the unified benchmark, and report our findings: (1) RDBMSs are significantly faster for aggregations and order by operations, (2) GDBMSs are shown to be superior for projection, multi-table join and deep recursive operations, (3) GDBMSs and RDBMSs are comparable for two-table join and shallow recursive operations.