Speculative query execution in RDBMS based on analysis of query stream multigraphs

Abstract

The paper presents an insight into a speculative execution model of queries in RDBMS based on the analysis of the stream of current queries appearing at the database input. A specific multigraph representation of input query stream is created and used to determine the speculative queries for execution. A group of worker threads execute the chosen speculative queries in parallel with the execution of the standard input stream of user queries. The obtained speculative results are then used to support faster query execution. First, the paper briefly reminds the assumed graph modelling and analysis methods. Then, additional rules are presented which enable combining results of multiple speculative queries in execution of a single user input query. The quality of executed and used speculations is then analysed based on the defined quality metrics and structural details of speculative queries. Conclusions from this analysis are used to modify the selection method of target queries for speculative execution. It aims at intensification of the use of multiple speculative query results and further reduction of the user query execution time. Experimental results are presented in a multi-threaded speculative experimental environment cooperating with a SQLite database. They show that with the improved algorithm we can obtain more varied speculative query results, and thus, more intensive use of multiple speculative query results by the stream of user queries sent to the database.