Evaluate using Profile-Guided Optimization (PGO)

[zamazan4ik]

Hi!

Several days ago I found the Reddit post about Optiprism and its technological stack.

I think applying PGO to Optiprism can help improve its performance and TCO (Total Cost of Ownership) in real production environments (as it already help other projects). According to the Reddit post above, there are already benchmarks in the awesome-pgo repo about optimizing various databases (including RocksDB) and similar software before and after PGO.

I suggest at first simply test the PGO approach with Optiprism. If it shows improvements - at least add a note in the documentation about optimizing Optiprism with PGO. Lately, we can try to optimize prebuilt binaries with PGO.

Thank you

[ravlio]

Great topic, thanks. I hope I'll find time to experiment with such optimizations. For now, there are a lot of things that can be improved just by fixing the code and it can gain x2-x10 performance :)

[zamazan4ik]

For now, there are a lot of things that can be improved just by fixing the code and it can gain x2-x10 performance :)

A nice thing about PGO is that it's still helpful even after "fixing the code" in the most bottlenecked places since PGO allows to the compiler perform better low-level code generation. Humans do high-level clever stuff, the compiler does the dirty low-level things.