Runtime Code Generation in Cloudera Impala.
Skye Wanderman-Milne, Nong Li · IEEE Data(base) Engineering Bulletin · 2014
In this paper we discuss how runtime code generation can be used in SQL engines to achieve better query execution times. Code generation allows query-specific information known only at runtime, such as column types and expression operators, to be used in performance-critical functions as if they were available at compile time, yielding more efficient implementations. We present Cloudera Impala, an open-source, MPP database built for Hadoop, which uses code generation to achieve up to 5x speedups in query times.