A simplified Architecture for Fast, Adaptive Compilation and Execution of SQL Queries

Abstract

Query compilation is crucial to efficiently execute query plans. In the past decade, we have witnessed considerable progress in this field, including compilation with LLVM, adaptively switching from interpretation to compiled code, as well as adaptively switching from non-optimized to optimized code. All of these ideas aim to reduce latency and/or increase throughput. However, these approaches require immense engineering effort, a considerable part of which includes reengineering very fundamental techniques from the compiler construction community, like register allocation or machine code generation – techniques studied in this field for decades. In this paper, we argue