Speculative execution plan for multiple query execution systems

Abstract

There are different levels at which parallelism can be introduced to the database system. Starting from data partitioning (intra-operator parallelism) up to parallelism of operation (inter-operator parallelism) that depends on a query granularity. The paper presents the parallelisation method based on speculative execution for the database systems which are expected to give answers to complex queries coming from different sources as soon as possible. Taking under consideration W of upcoming queries waiting for execution, the execution plan for the first query should be developed. This plan should give the largest benefit also for W-1 of the consecutive queries. Thus, in parallel to the first query, some excessive computations can be executed, which in further steps would reduce the execution time of the consecutive queries. The paper presents possible risks and benefits are using this method and also analyses of possible execution time reduction for different models of speculative parallelization [1].