Graph DBs vs. Column-Oriented Stores: A Pure Performance Comparison

Cloud Computing has brought a great change in the way information is stored and applications run. In order for one or more clusters to work as a cloud we need a middleware framework, such as Apache Hadoop [17], that provides reliability, scalability and distributed computing. Once the infrastructure has been established, a software framework can be installed, which runs on top of it and will be the connection to communicate with the applications developed by the users. The software, in this regard, is a NoSQL database. This paper deals with the problem of searching data in some widespread NoSQL databases used in cloud computing. Two categories of NoSQL databases are compared; one based on columns using a column-oriented key-value store, HBase [6], and a high-available graph database, Neo4j [11]. HBase is a distributed, scalable storage system that runs on top of HDFS, and has being designed based on Google’s BigTable [4]. Neo4j has being designed and developed to be a reliable database, optimized for graph structures, instead of tables, and is a robust, scalable, high performance and high available database that supports ACID transactions and queries written in Cypher language. The aim of this paper is to create a novel system that will decide when a query must be send to be executed in a key-value store or a graph database. Thus, an experimental pure performance comparison has been made between Apache HBase and Neo4j for a variety of queries, that were programmed using systems API’s and Java language.