Stefan Zeiger

A lot of preliminary work was done for soft inserts and insertAndUpdate support in 2.0 and 2.1. It should be possible for updates to use the same mechanism as for...

For reference, this is the raw data I collected: ``` 17.0.11-graal: [info] welcome to sbt 1.9.9 (Oracle Corporation Java 17.0.11) [success] Total time: 0 s, completed Jul 8, 2024, 4:28:53...

And this is the data from our internal benchmark, building part of our monorepo with our custom Bazel-based Scala tooling: - Baseline: Most of the compilation on Zulu 8, with...

Your results are unexpected. The first one or two runs being slower is normal due to the warmup, but there shouldn't be a big variance afterwards. I think a 2x...

@c-refice How do I make async-profiler not write anything to stdout (which is always a PITA with Java tooling)? We're compiling with a Bazel worker, so any additional output must...

It looks like I can run asprof manually to avoid any output from the JVM. I'll do some test runs and collect the data, and then try to find a...

I did 3 runs each on 17 and 21: ``` ~/tmp/profile_17ee_1.jfr: Elapsed time: 559.912s, Critical Path: 516.60s * ~/tmp/profile_17ee_2.jfr: Elapsed time: 578.837s, Critical Path: 504.14s ~/tmp/profile_17ee_3.jfr: Elapsed time: 568.723s, Critical...

@c-refice I ran another set of benchmarks with a smaller build. The difference in build time is also comparatively smaller. Perhaps it doesn't affect all projects equally. https://drive.google.com/file/d/1Smc_zclTADQhY0TYURnpFCtBtv45GuAF/view?usp=sharing contains 6...

done

I've done some testing on GraalVM 24 EE. Performance for Scala compilation is hugely improved. Initial benchmarks with a large, highly-parallel build (in a single JVM on EC2 m7i.24xlarge) show:...