Cannot build with Java 1.8.0_11

[llogiq]

Hi there,

I'm currently using trove for a few projects, so naturally I wanted to benchmark hftc in comparison. However, the build buildMain always stops at > Building 8% > :lib:impl:compileJava.

My setup is Java 1.8.0_11 (oracle) on LUbuntu 13.10 (64bit), gradle 2.0, groovy 2.3.3, ant 1.9.3.

PS. It would be great if you could release binaries.

[leventov]

HFTC build is very time & memory consuming. On my not so week dev machine (8 GB RAM, SSD) the full build takes more than half an hour. I have never tried to run it on HDD. :lib:impl:compileJava takes the most memory among all tasks, more than 3 GB.

If you build yourself, it seems practical to generate less specializations, excluding which you don't need. for example, add these lines after https://github.com/OpenHFT/hftc/blob/e64f5e34bc7d02738cff143dcab35cca3e759033/lib/build.gradle#L31:

    never 'byte|char|short'
    exclude 'key=float|double,value=float|double|int|long|obj'
    exclude 'key=int,value=obj|float'

Also, why you can't get these binaries: https://oss.sonatype.org/content/groups/public/net/openhft/hftc-api-jdk8/0.5/hftc-api-jdk8-0.5.jar and https://oss.sonatype.org/content/groups/public/net/openhft/hftc-impl-jdk8/0.5/hftc-impl-jdk8-0.5.jar?

[llogiq]

Thanks - I wouldn't have found those binaries without your link; you may consider linking them from README.md for us Sonatype-impaired folks?

I took the binaries and ran a small test for IntHashSet (writing 50k random numbers, checking 100k random numbers, some of which are probably not within the set).

I may have too puny a machine (only 4GB of ram, sluggish HDD), but hftc ran roughly 20-40× slower than a comparable trove TIntHashSet. I figure I must be doing something wrong, though.

[leventov]

Could you show benchmarking code, please?

[llogiq]

Here you are.

Yes, I know I should have used JMH instead of caliper, but somehow my mvn got borked, thus JMH doesn't run ATM. Still, the results are so far off that I think there must be something amiss & it should be easy enough to reproduce.

Regards, Andre

2014-07-28 13:57 GMT+02:00 Roman Leventov notifications@github.com:

Could you show benchmarking code, please?

— Reply to this email directly or view it on GitHub https://github.com/OpenHFT/hftc/issues/24#issuecomment-50329307.

[leventov]

Sorry, I don't see the code or the link to.

I'm sure Caliper is good enough for this type of benchmark. But it is anyway interesting to explore the case.

[llogiq]

Ok, I've put the code into a public gist

[leventov]

@llogiq, that isn't a problem of your bench, that was an awful bug in HFTC implementation :( That's surprising how could it live in the repo for so long time.

Now results are (I rewritten in JMH without changes, because there are problems with running Caliper on Windows):

Benchmark                             Mode   Samples         Mean   Mean error    Units
t.HftcVsTroveIntSet.timeHftcHash      avgt        10       29,173        0,632    ns/op
t.HftcVsTroveIntSet.timeTroveHash     avgt        10       56,952        0,121    ns/op

I have released new binaries with a fix: https://oss.sonatype.org/content/groups/public/net/openhft/hftc-api-jdk8/0.5.1/hftc-api-jdk8-0.5.1.jar and https://oss.sonatype.org/content/groups/public/net/openhft/hftc-impl-jdk8/0.5.1/hftc-impl-jdk8-0.5.1.jar

[leventov]

@llogiq, thank you very much.

[llogiq]

You're welcome. Thanks for the timely fix and keep up the good work!