Profile-Guided Optimization (PGO) performance results

Hi!

Recently I did many Profile-Guided Optimization (PGO) benchmarks on multiple projects - the results are available here. I think since the project is performance-oriented, it would be interesting to try to test PGO for optimizing tquic. I already did some benchmarks.

Test environment

Fedora 38
Linux kernel 6.5.5
AMD Ryzen 9 5900x
48 Gib RAM
SSD Samsung 980 Pro 2 Tib
Compiler - Rustc 1.73
tquic version: the latest for now from the develop branch on commit 05c56e7425ec1149a9c95ca7bbcb6acbab861fd6
Disabled Turbo boost

Benchmark setup

For benchmarking purposes, I use the project's benchmarks. Release benchmarking is done with cargo bench, PGO optimized build is done with cargo-pgo with cargo pgo bench && cargo pgo optimize bench. PGO profiles are collected from the benchmark workload itself.

Results

I got the following results:

Release + PGO-optimized compared to Release: https://gist.github.com/zamazan4ik/875e3d1948ac3db575bcbb442a6f6ebb
(just for reference) PGO-Instrumented compared to Release: https://gist.github.com/zamazan4ik/f4af4f2e510d433253bf007ecb584507

According to the tests, PGO consistently improves tquic performance in some scenarios.

Further steps

I can suggest the following things to do:

Evaluate PGO's applicability to tquic in more scenarios.
If PGO helps to achieve better performance - add a note to tquic's documentation about that (probably somewhere in the README file). In this case, users and maintainers will be aware of another optimization opportunity for tquic.
Maybe get some insights from the PGO profiles and optimize manually the code according to the profiles (maybe more aggressive inlining or something like that)

Tencent / tquic