Graph-based speculative query execution for RDBMS

Abstract

The paper concerns parallelized speculative support for query execution in RDBMS. The support is based on dynamic analysis of input query stream in databases serviced in SQLite. A multi-threaded middleware called the Speculative Layer is introduced which, based on a specific graph representation of a query stream, determines the most promising speculative queries for execution. The paper briefly presents the query graph modelling and analysis methods. Then, an extended version of speculative query execution algorithm is presented which allows combining results of multiple speculative queries for execution of one user input query. Experimental results are presented based on the proposed algorithm assessment in a multi-threaded testbed cooperating with a database serviced in SQLite.