TGDB: towards a benchmark for graph databases

Graph data has become an important representation for many analytical applications, ranging from social network analysis to biological data computation, to ontologies in the semantic web. Recently, many graph databases have been proposed to process and analyze graph data. We can categorize these into two main approaches: one is to build a layer of graph data model on top of an existing database (e.g., key-value store); and the second is to build a specialized native data processing substrate for processing graph data. Consequently, data scientists at present have a variety of choices and approaches to choose amongst. This requires having an approach to evaluate and assess these approaches, to select the one that suits best their situation. We propose TGDB, the Toronto Graph Database Benchmark. TGDB has query workload and real-world datasets to evaluate the performance of targeted systems. We choose three graph databases that have different system architectures and evaluate their performance against TGDB.