CONST: Continuous online NoSQL schema tuning

Abstract

AbstractSelf‐tuning is a feature of autonomic databases that includes the problem of automatic schema design. Automating the schema design process is critical to provide an optimized schema that increases the performance of an application's workload in NoSQL databases. There has been important work on automated design tuning for NoSQL databases. Existing solutions take an offline approach to the schema design problem and leave several important decisions to database administrators. Although offline approaches recommend good and efficient schemas, their analysis is far too heavy‐weight to be run continuously. This paper develops a novel self‐tuning control loop, called CONST, for online monitoring, analysis, and schema tuning a managed NoSQL database system that automatically modify the current schema design reacting to changes in the application's workload. CONST provides components that fulfill IBM's MAPE‐K loop phases to support self‐tuning. We describe an implementation of the proposed CONST loop on top of the Cassandra wide column store and evaluate its performance experimentally. Our results show the low overhead of CONST and demonstrate its ability to tune a schema design in response to changes in the application's workload.