Everything You Always Wanted to Know About Compiled and Vectorized Queries But Were Afraid to Ask 论文阅读笔记
看了 Ref[1] 和 Ref[2]，基本就差不多了
Vectorization：materialization 开销，可以利用 SIMD 并行数据操作，最好是 column store
Code gen：指令数少，利于计算密集型
- join (memory bound)：向量化快
- memory load 消耗 CPU cycle，向量化减少 cache miss
- computation (CPU intensive task)：code gen 快
- cache 压力小，code gen 指令数少，高效利用 register
- selection 使用 SIMD
- 越多 select，越稀疏，column 上 offset 越大，导致 cache miss
消除分支的操作：a>b?1:0 可以被写成没有分支的语句 setg
Reference
- 在 2019.4.20 杭州举办的 Infra Meetup No.98 上，我司 TiDB 研发工程师徐怀宇为大家带来了《Everything You Always Wanted to Know About Compiled and Vectorized Queries But Were Afraid to Ask》论文分享。
- CMU 15-721 Advanced Database Systems (Spring 2018)
- A Deep Dive into Query Execution Engine of Spark SQL - Maryann Xue - You Tube
- Vectorization vs. compilation in query execution