RFR: 8316585: [REDO] runtime/InvocationTests spend a lot of time on dependency verification

Aleksey Shipilev shade at openjdk.org
Thu Oct 12 14:29:59 UTC 2023


On Thu, 5 Oct 2023 10:28:48 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> This is another (hopefully better) attempt at fixing the issue described by [JDK-8315932](https://bugs.openjdk.org/browse/JDK-8315932). The previous attempt tried to tune the Metaspace parameters, hoping for a more frequent GCs. This patch just does the GCs explicitly in the tests.
> 
> There is a tradeoff between spending more on frequent GCs and having fewer dependencies to verify. But it is a win on all thresholds. I have selected the best threshold with a light bisection.
> 
> 
> # Testing with CONF=linux-x86_64-server-fastdebug make test TEST=runtime/InvocationTests/  
> 
> # Current run time
> 4312.23s user 93.50s system 467% cpu 15:42.02 total
> 
> Original fix:
> 2875.61s user 116.28s system 677% cpu 7:21.57 total
> 
> # GC every 8192 classes
> 3666.35s user 98.30s system 489% cpu 12:49.13 total
> 
> # GC every 4096 classes
> 3083.87s user 104.52s system 586% cpu 9:04.06 total
> 
> # GC every 3000 classes <---- this PR
> 3066.15s user 110.85s system 613% cpu 8:37.96 total
> 
> # GC every 2500 classes
> 3079.33s user 114.35s system 582% cpu 9:07.85 total
> 
> # GC every 2048 classes
> 3171.62s user 116.97s system 492% cpu 11:07.38 total
> 
> # GC every 1024 classes
> 3112.66s user 114.94s system 506% cpu 10:37.35 total
> 
> # GC every 512 classes
> 3180.21s user 122.11s system 508% cpu 10:48.85 total
> 
> # GC every 128 classes
> 3628.81s user 90.54s system 481% cpu 12:52.88 total

Anyone? Particularly @tstuefe, @iwanowww?

-------------

PR Comment: https://git.openjdk.org/jdk/pull/16052#issuecomment-1759719536


More information about the hotspot-runtime-dev mailing list