Is your feature request related to a problem? Please describe.
Not a problem - just an opportunity.
Describe the solution you'd like
Profile-Guided Optimization could help with gaining more performance with Quickwit. Would be great to test PGO on Quickwit on near real-world test cases. Probably something from https://github.com/quickwit-oss/quickwit/issues/794 could be used.
Additional context
I did a quick research locally. On the latest master branch (version 0.6.1) and using for training and evaluation this benchmark, I didn't get a performance improvement (PGO version works at the same speed as non-PGO release build) but reduced a binary size: from 92 Mib to 85 Mib. My guess is this is due to less aggressive inlining since the benchmark executes only a few possible hot paths.
fulmicoton
commented
1 year ago
Is your feature request related to a problem? Please describe. Not a problem - just an opportunity.
Describe the solution you'd like Profile-Guided Optimization could help with gaining more performance with Quickwit. Would be great to test PGO on Quickwit on near real-world test cases. Probably something from https://github.com/quickwit-oss/quickwit/issues/794 could be used.
Additional context I did a quick research locally. On the latest master branch (version 0.6.1) and using for training and evaluation this benchmark, I didn't get a performance improvement (PGO version works at the same speed as non-PGO release build) but reduced a binary size: from 92 Mib to 85 Mib. My guess is this is due to less aggressive inlining since the benchmark executes only a few possible hot paths.
I think with a more mature benchmark something like this https://github.com/ClickHouse/ClickBench we could try to speed up Quickwit.
More info about PGO you could find here: https://github.com/ZaMaZaN4iK/awesome-pgo