Validate benchmarks

[skjolber]

The benchmarks are showing very good results, it would be interesting to get some more eyes on and also running them on more hardware.

[aroger-r7]

hey just came across your benchmark, had a quick look there is quite a lot.
first thing is that you might want to upgrade sfm to version 6.0.9.

[skjolber]

@aroger-r7 Right, bumped, should the code be adjusted as well as the dependency?

[aroger-r7]

there CsvReader has moved package but I think that should be it

[arnaudroger]

so run it locally with better -i 5 -wi 5 and f 5 - took more than 24 hours. but you mapper is faster. I wonder if that comes down to the number parsing. unfortunately, the gradle plugin does not build a shaded jar. but I'll try to get some profiling info to see.

[skjolber]

Uh yeah, the run time should be adjusted, kind of long now.

Edit: 👍 I read that a bit too fast it seems

[skjolber]

There is actually two simpleflatmapper dependencies, the other is in https://github.com/skjolber/csv-benchmark/blob/master/jmh-benchmark/build.gradle and I forgot to bump it the first time around. Edit: So that might have skewed the results. Have not had the time to run new ones yet.

[skjolber]

@arnaudroger I see the number parsing has been improved in sfm - did that affect the benchmarking results?

[arnaudroger]

slightly, but your mapper still faster. got a bit of improvement by doing a parseInt on char[] directly but still quite far :)

…

On Tue, Nov 27, 2018 at 5:25 PM Thomas Skjølberg ***@***.***> wrote: @arnaudroger <https://github.com/arnaudroger> I see the number parsing has been improved in sfm - did that affect the benchmarking results? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF4Erbh-NkPkA3J_RQaVazjILdaIbyufks5uzXWFgaJpZM4WfiDS> .

[arnaudroger]

just pushed 6.0.10 with some pref improvement for sfm-csv. still not as fast :) not that surprising. I don't think I can get it as low because the support for constructor injection, sub object and other yaml comment.

[skjolber]

I'll refresh the benchmark results anyhow (within 2-3 weeks).

[skjolber]

@arnaudroger Benchmarks refreshed. My gut feeling is that SFM improved somewhat.