[JBTM-4014] Reduce allocation pressure on Uid #2414

marcosgopen · 2025-10-14T14:22:30Z

CORE AS_TESTS RTS !JACOCO XTS QA_JTA QA_JTS_OPENJDKORB PERFORMANCE DB_TESTS

Improving Uid and utility performance

jbosstm-bot · 2025-10-14T14:23:45Z

Started testing this pull request with PERF profile: https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=PERFORMANCE,jdk=openJDK17,label=jnlp-agent/8/

jbosstm-bot · 2025-10-14T14:24:11Z

Started testing this pull request with PERF profile: https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=PERFORMANCE,jdk=openJDK21,label=jnlp-agent/8/

jbosstm-bot · 2025-10-14T15:40:03Z

Benchmark output (please refer to the article https://developer.jboss.org/wiki/PerformanceGatesForAcceptingPerformanceFixesInNarayana for information on our testing procedures.

If you just want to run a single benchmark then please refer to the README.md file in our benchmark repository at https://github.com/jbosstm/performance/tree/main/narayana

For information on the hardware config used for this PR please consult the CI job artefact hwinfo.txt or the job output

jbosstm-bot · 2025-10-14T15:40:03Z

Benchmark output (please refer to the article https://developer.jboss.org/wiki/PerformanceGatesForAcceptingPerformanceFixesInNarayana for information on our testing procedures.

If you just want to run a single benchmark then please refer to the README.md file in our benchmark repository at https://github.com/jbosstm/performance/tree/main/narayana

For information on the hardware config used for this PR please consult the CI job artefact hwinfo.txt or the job output

jbosstm-bot · 2025-10-14T15:40:04Z

PERFORMANCE profile job finished https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=PERFORMANCE,jdk=openJDK17,label=jnlp-agent/8/

jbosstm-bot · 2025-10-14T15:40:04Z

PERFORMANCE profile job finished https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=PERFORMANCE,jdk=openJDK21,label=jnlp-agent/8/

franz1981 · 2025-10-16T11:34:21Z

@marcosgopen Anything more I should do here to help? 🙏
Thanks again for opening it, I was a bit scared of the codestyle :D

marcosgopen · 2025-10-17T16:05:20Z

ArjunaCore/arjuna/classes/com/arjuna/ats/arjuna/common/Uid.java

+          pos += secChars;
+          asciiBytes[pos++] = (byte) Uid.breakChar;
+          Utility.toHexChars(other, asciiBytes, pos, otherChars);
+          _stringForm = new String(asciiBytes, 0, 0, totalChars);


@franz1981 I can see this constructor is deprecated, could you use another constructor please?

It's deprecated from more than 10 years (much more actually) and not planned (yet) to be removed
see https://docs.oracle.com/javase/1.5.0/docs/api/java/lang/String.html#String(byte[],%20int,%20int,%20int)
which is Java 5
Hotspot OpenJDK team is still working to make others as perfomant
see https://bugs.openjdk.org/browse/JDK-8295496 where I made the Performance OpenJDK team to work on it (actually I should work on it, but I have little time to fix it) - where there are benchmarks for the other versions.
Which make me think that _stringForm = new String(asciiBytes, 0, 0, totalChars); should be repalced by
_stringForm = new String(asciiBytes, 0);

franz1981 · 2025-10-29T15:21:58Z

FYI franz1981@739493b
based on what observed on benchmarks ^^

jbosstm-bot · 2025-10-29T15:27:39Z

QA_JTS_OPENJDKORB profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=QA_JTS_OPENJDKORB,jdk=openJDK17,label=jnlp-agent/39/): Narayana rebase on main failed. Please rebase it manually

jbosstm-bot · 2025-10-29T15:27:53Z

PERFORMANCE profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=PERFORMANCE,jdk=openJDK17,label=jnlp-agent/39/): Narayana rebase on main failed. Please rebase it manually

jbosstm-bot · 2025-10-29T15:27:56Z

Tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=postgres,jdk=openJDK17,label=jnlp-agent/65/): Narayana rebase on main failed. Please rebase it manually

jbosstm-bot · 2025-10-29T15:28:08Z

QA_JTS_OPENJDKORB profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=QA_JTS_OPENJDKORB,jdk=openJDK21,label=jnlp-agent/39/): Narayana rebase on main failed. Please rebase it manually

jbosstm-bot · 2025-10-29T15:28:10Z

XTS profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=XTS,jdk=openJDK17,label=jnlp-agent/39/): Narayana rebase on main failed. Please rebase it manually

mmusgrov · 2025-10-31T10:04:57Z

Tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=postgres,jdk=openJDK17,label=jnlp-agent/65/): Narayana rebase on main failed. Please rebase it manually

@franz1981 will you rebase your PR please.

jbosstm-bot · 2025-10-31T15:09:06Z

Started testing this pull request with RTS profile: https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=RTS,jdk=openJDK17,label=jnlp-agent/41/

jbosstm-bot · 2025-10-31T15:17:24Z

Started testing this pull request with DB_TESTS profile: https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=postgres,jdk=openJDK17,label=jnlp-agent/66/

jbosstm-bot · 2025-10-31T15:18:37Z

Started testing this pull request with DB_TESTS profile: https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=oracle,jdk=openJDK17,label=jnlp-agent/66/

jbosstm-bot · 2025-10-31T15:22:43Z

DB_TESTS profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=postgres,jdk=openJDK17,label=jnlp-agent/66/): narayana build failed

jbosstm-bot · 2025-10-31T15:25:54Z

DB_TESTS profile tests failed (https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana-jdbcobjectstore/DB_TYPE=oracle,jdk=openJDK17,label=jnlp-agent/66/): narayana build failed

jbosstm-bot · 2025-10-31T16:08:05Z

RTS profile tests passed - Job complete https://jenkins-csb-narayana-ci.dno.corp.redhat.com/job/btny-pulls-narayana/PROFILE=RTS,jdk=openJDK17,label=jnlp-agent/41/

mmusgrov · 2025-10-31T16:21:23Z

@franz1981 Thanks for rebasing. We have been having Jenkins CI problems recently, please bear with us while we investigate.

franz1981 · 2025-11-01T11:54:56Z

I will play in the weekend with some branchless algorithm but will hurt badly the readability, for a much higher throughput. Although tbh I don't think this one to be that slow compared to the openJDK one but will happily measure it

franz1981 · 2025-11-01T13:42:10Z

Here we go, I have written it on a paper but not tried yet

private static long nibblesToAscii(long v) {
    long carry = ((v + 0x0606060606060606L) >>> 4) & 0x0101010101010101L;
    return v + 0x3030303030303030L + (carry * 0x27L);
}

This is fixing the encoded hex for nibbles mapping to a-f with the carry (which is 0 if 0-9 and 1 if a-f) multiplying per 0x27 the carry to produce a batch adjustment.
This approach has the downside of a multiplication but produce a single batch encoded hex which can be written in one go via VarHandle.
It cannot still be used as it is because we start with 8 bytes read and it could produce 16 ASCII hex chars, so we need a way to split the 8 bytes in low/hi nibbles, to use any of these methods here.
Clearly it is less maintainable but can be highly effective.
I will play with some benchmark for fun as I have some time.

And this one is saving the multiplication, too:

private static long nibblesToAscii(long v) {
    // For both cases: result = v + 0x30 + (v >= 10 ? 0x27 : 0)
    // But we can compute (v >= 10) as (v + 6) >>> 4 & 1
    long base = v + 0x3030303030303030L;
    long adjustment = ((v + 0x0606060606060606L) >>> 4) & 0x2727272727272727L;
    
    return base + adjustment;
}

franz1981 · 2025-11-02T09:27:07Z

Thinking about it twice since most of time we rarely encode the full 64 bit length is not worthy to have a swar variant

mmusgrov · 2025-11-03T08:54:08Z

not worthy to have a swar variant

What's a swar variant?

franz1981 · 2025-11-03T09:06:38Z

The one based on #2414 (comment)

Which can process batches of bytes at a time

marcosgopen · 2025-11-03T14:49:47Z

Thanks @franz1981 for updating your branch! I initially opened this PR to test your improvements but I think it makes more sense if we close this PR and we move the conversation to a new PR (same branch) opened by the branch's owner. Thanks again

franz1981 · 2025-11-03T15:02:08Z

@marcosgopen this is already from my branch

marcosgopen · 2025-11-03T15:11:21Z

@marcosgopen this is already from my branch

Sorry, I can clarify: if you can open a new PR with this changes I can review it too. Otherwise I might seem the author of this PR's changes. Thanks

franz1981 · 2025-11-04T03:48:59Z

No worries @marcosgopen

Just a qq, I cannot see unit tests for Uid; I would like to write the one because the main difference with the new vs old way to encode hex is that I'm considering all long/int as unsigned - but Uid::minUid is actually negative, which means that I have to modify the util methods I created to work well with that

marcosgopen · 2025-11-04T09:14:27Z

Closing in favor of #2427

marcosgopen · 2025-11-04T09:16:39Z

No worries @marcosgopen

Just a qq, I cannot see unit tests for Uid; I would like to write the one because the main difference with the new vs old way to encode hex is that I'm considering all long/int as unsigned - but Uid::minUid is actually negative, which means that I have to modify the util methods I created to work well with that

@franz1981 you can find it here: https://github.com/jbosstm/narayana/blob/main/ArjunaCore/arjuna/tests/classes/com/hp/mwtests/ts/arjuna/uid/UidUnitTest.java

marcosgopen marked this pull request as ready for review October 17, 2025 15:57

marcosgopen commented Oct 17, 2025

View reviewed changes

franz1981 mentioned this pull request Oct 30, 2025

JBTM-4014-perf-test draft jbosstm/performance#185

Open

franz1981 added 2 commits October 31, 2025 15:59

[JBTM-4014] Reduce allocation pressure on Uid

256abf1

Optimize stringForm to use file break char, if needed

318f6df

franz1981 force-pushed the master branch from 739493b to 318f6df Compare October 31, 2025 15:04

marcosgopen closed this Nov 4, 2025

[JBTM-4014] Reduce allocation pressure on Uid #2414

[JBTM-4014] Reduce allocation pressure on Uid #2414

Uh oh!

Conversation

marcosgopen commented Oct 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

jbosstm-bot commented Oct 14, 2025

Uh oh!

franz1981 commented Oct 16, 2025

Uh oh!

marcosgopen Oct 17, 2025

Choose a reason for hiding this comment

Uh oh!

franz1981 Oct 17, 2025

Choose a reason for hiding this comment

Uh oh!

franz1981 commented Oct 29, 2025

Uh oh!

jbosstm-bot commented Oct 29, 2025

Uh oh!

jbosstm-bot commented Oct 29, 2025

Uh oh!

jbosstm-bot commented Oct 29, 2025

Uh oh!

jbosstm-bot commented Oct 29, 2025

Uh oh!

jbosstm-bot commented Oct 29, 2025

Uh oh!

mmusgrov commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

jbosstm-bot commented Oct 31, 2025

Uh oh!

mmusgrov commented Oct 31, 2025

Uh oh!

franz1981 commented Nov 1, 2025

Uh oh!

franz1981 commented Nov 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

franz1981 commented Nov 2, 2025

Uh oh!

mmusgrov commented Nov 3, 2025

Uh oh!

franz1981 commented Nov 3, 2025

Uh oh!

marcosgopen commented Nov 3, 2025

Uh oh!

franz1981 commented Nov 3, 2025

Uh oh!

marcosgopen commented Nov 3, 2025

Uh oh!

franz1981 commented Nov 4, 2025

Uh oh!

marcosgopen commented Nov 4, 2025

Uh oh!

marcosgopen commented Nov 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

marcosgopen commented Oct 14, 2025 •

edited

Loading

franz1981 commented Nov 1, 2025 •

edited

Loading

marcosgopen commented Nov 4, 2025 •

edited

Loading