Using Profile-Guided Optimization (PGO) to reduce the encode overhead even more

[zamazan4ik]

Hi!

Similarly to vincent-herlemont/native_db#92 I decided to perform PGO benchmarks on native_model. Here are the results.

Test environment

• Fedora 39
• Linux kernel 6.6.9
• AMD Ryzen 9 5900x
• 48 Gib RAM
• SSD Samsung 980 Pro 2 Tib
• Compiler - Rustc 1.75
• native_model version: the latest for now from the main branch on commit 62b1e7cc35e64bce9feb22d6727a4f66fc9b9660
• Disabled Turbo boost (for more stable results across benchmark runs)

Benchmark

For benchmark purposes, I use built-in benchmarks with cargo bench command. For PGO optimization I use cargo-pgo tool. The same benchmark suite was used for the PGO training phase built with cargo pgo bench. PGO optimized results I got with cargo pgo optimize bench.

All measurements are done multiple times to check reproducibility - the results are stable across runs.

Results

I got the following results:

• Release: https://gist.github.com/zamazan4ik/1696f6135bcb6d69250cee5ec5079619
• PGO optimized compared to Release: https://gist.github.com/zamazan4ik/5c876ec7f3f326735cfcfb13af25828d
• (just for reference) PGO instrumentation compared to Release: https://gist.github.com/zamazan4ik/be67cec26d896461ad52b6f99c4e91e9

At least according to the results above, PGO helps with achieving better overall performance with native_model. Probably the PGO-optimized build can suggest a way how to optimize native_model more aggressively (via comparing ASM for PGOed and non-PGOed native_model versions).

Further steps

I can suggest the following action points:

• Perform more PGO benchmarks on native_model. If it shows improvements - add a note to the documentation about possible improvements in native_model performance with PGO. So native_model users will be aware of PGO effects on native_model performance and can decide to enable PGO for their native_model-based applications to achieve better performance.

Please treat the issue just as a benchmark report, not a problem or something like that. I created the issue just because the discussions are not enabled in this repo.