Enabling JSON Document Stores in Relational Systems

In recent years, “document store” NoSQL systems have exploded in popularity. A large part of this popularity has been driven by the adoption of the JSON data model in these NoSQL systems. JSON is a simple but expressive data model that is used in many Web 2.0 applications, and maps naturally to the native data types of many modern programming languages (e.g. Javascript). The advantages of these NoSQL document store systems (like MongoDB and CouchDB) are tempered by a lack of traditional RDBMS features, notably a sophisticated declarative query language, rich native query processing constructs (e.g. joins), and transaction management providing ACID safety guarantees. In this paper, we investigate whether the advantages of the JSON data model can be added to RDBMSs, gaining some of the traditional benefits of relational systems in the bargain. We present Argo, an automated mapping layer for storing and querying JSON data in a relational system, and NoBench, a benchmark suite that evaluates the performance of several classes of queries over JSON data in NoSQL and SQL databases. Our results point to directions of how one can marry the best of both worlds, namely combining the flexibility of JSON to support the popular document store model with the rich query processing and transactional properties that are offered by traditional relational DBMSs.