Using SIMD to aggregate billions of rows per second - QuestDB

Using SIMD to aggregate billions of rows per second - QuestDB

SIMD instructions are specific CPU instruction sets for arithmetic calculations that use synthetic parallelisation. The parallelisation is synthetic because instead of spreading the work across CPU cores, SIMD performs vector operations on multiple items using a single CPU instruction. In practice, if you were to add 8 numbers together, SIMD does that in 1 operation instead of 8. We get compounded performance improvements by combining SIMD with actual parallelisation and spanning the work across CPUs.

Read more...

На русском...

要查看或添加评论,请登录

Igor Suhorukov的更多文章

社区洞察

其他会员也浏览了