Performance Evaluations of Document-Oriented Databases Using GPU and Cache Structure

Document-oriented databases are popular databases, in which users can store their documents in a schema-less manner and perform search queries for them. They have been widely used for web applications that process a large collection of documents because of their high scalability and rich functions. One of major functions of document-oriented databases is a string search that requires a high computational cost for a large collection of documents, because its computational complexity increases as the documents increase. In document-oriented databases, a database index is typically used for improving text search queries. However, the index cannot always be used for text search queries, such as a regular expression match search. To accelerate such queries by using GPUs, in this paper, we propose a GPU-friendly cache structure, called DDB Cache (Document-oriented DataBase Cache), which is extracted from a document-oriented database. By using GPU and DDB Cache, we can improve a performance of text search queries without relying on the database indexes. We implemented DDB Cache for MongoDB. Experimental results using GeForce GTX 980 show that our approach improves the performance of regular expression search queries by up to 101x compared to the original document-oriented database.