Canfor
50-Minute Talk
Compilers and modern CPUs go to immense efforts make code run fast. For programmers, even code that at first glance may appear quite optimal can often contain many sub-optimal aspects that could be changed to help give the compiler and CPU the things they need to make the code run faster.
In this talk, we'll go through some examples of what I discovered on a recent journey to make tuple deformation faster in PostgreSQL. There were many independent improvements made, and I'd like to talk about these, as there might be other places where similar changes could be made. For each change, we'll go into detail about why that change was made and talk about why this helped performance. When relevant, we'll go through a CPU profiling demonstration to show how the given change helps. In other cases, benchmark results will be shown to demonstrate that the change had an impact on performance and by how much.
Finally, I will talk about what could be done in the future to further speed up tuple deformation, and I'll also leave some tips for people who would like to go on a similar journey.
