A Simple Approach for Executing SQL on a NoSQL Datastore

NoSQL datastores have been initially introduced to support a few concrete extreme scale applications. Limited query and indexing capabilities were therefore not a major impediment, as the specificity and scale of the target application justified the investment in manually crafting application code. With a number of alternatives now available and mature, there is an increasing willingness to use NoSQL datastores in a wider and more diverse spectrum of applications, where hand-crafted query code is a much less enticing trade-off. We address this challenge with a simple approach for running SQL queries on top of a NoSQL datastore while preserving the underlying scalability, flexible schema and transaction-less semantics. We demonstrate our approach with a running prototype atop HBase. Our evaluation, conducted using YCSB shows the minimal overhead compared to the direct usage of HBase and the evaluation using an unmodified SQL implementation of a standard relational database workload, TPC-C, shows that the proposal presents linear scalability. Moreover, the comparison with a TPC-C implementation optimized for HBase, shows that complex SQL applications can be easily run and even achieve better results.