Switch to Eclipse OpenJ9 (to gain performance)

[koppor]

Background and Motivation

Eclipse OpenJ9 is a JVM developed by IBM since the 90s. It has superior performance compared to Oracle's JVM: https://www.eclipse.org/openj9/performance/

We should give it a try in JabRef, too.

TODOs

• Internal PR Switch to Eclipse OpenJ9 JDK v15 jabref#7280 until Unhandled exception during compilation with JDK 15.0.1.j9 eclipse-openj9/openj9#11549 is solved.

Changes

• Removes jdk.jfr as this is unsupported in OpenJ9 - JEP 328 Flight Recorder eclipse-openj9/openj9#1876

---

• Change in CHANGELOG.md described (if applicable)
• Tests created for changes (if applicable)
• Manually tested changed features in running JabRef (always required)
• Screenshots added in PR description (for UI changes)
• Checked documentation: Is the information available and up to date? If not created an issue at https://github.com/JabRef/user-documentation/issues or, even better, submitted a pull request to the documentation repository.

[koppor]

• Blocked by Pure Java implementation michel-kraemer/citeproc-java#92, because of 14.0.2.12-openj9 fails with module not found: jdk.internal.vm.compiler eclipse-openj9/openj9#11555.

[koppor]

• Now blocked by Unhandled exception during compilation with JDK 15.0.1.j9 eclipse-openj9/openj9#11549

[Siedlerchr]

As we are now at java 16 this can be closed right?

[koppor]

As we are now at java 16 this can be closed right?

No. The first two paragraphs of the PR description still remain valid:

Eclipse OpenJ9 is a JVM developed by IBM since the 90s. It has superior performance compared to Oracle's JVM: https://www.eclipse.org/openj9/performance/

We should give it a try in JabRef, too.

[koppor]

Blocked by:

D:\a\jabref\jabref\build\jlinkbase\tmpjars\org.jabref.merged.module\module-info.java:1446: error: module not found: jdk.jfr
    requires jdk.jfr;
                ^
1 error

• Update: Should be fixed by Result of ./gradlew suggestMergedModuleInfo jabref#7713
• Update: OpenJ9 does not support Java Flight Recorder - we do not use it on the distribution, thus a removal is OK -- JEP 328 Flight Recorder eclipse-openj9/openj9#1876

[koppor]

OpenJ9is now officially supported by setup-java: https://github.com/actions/setup-java#supported-distributions

[koppor]

Blocked by

Error: jdk.internal.org.objectweb.asm.MethodTooLargeException: Method too large: jdk/internal/module/SystemModules$all.moduleDescriptors ()[Ljava/lang/module/ModuleDescriptor;

• Refs JDK-8240567: MethodTooLargeException thrown while creating a jlink image: [ERROR] Error: jdk.internal.org.objectweb.asm.MethodTooLargeException: Method too large
• Refs Issues depending on external libraries jabref#5226

[koppor]

Google found https://bell-sw.com/announcements/2022/06/28/hotspot-vs-openj9-performance-comparison/

The experiment results are similar on all tested platforms and demonstrate that Liberica JDK with
configured HotSpot is comparable to or better than OpenJ9.
t is worth noting that the default HotSpot gives much lower latency and higher throughput than OpenJ9 on a server-class node

However, OpenJ9 can be configured even more, which was not done by the Liberica-Persons. See https://www.eclipse.org/openj9/docs/openj9_newuser/.

[koppor]

Other comment - from https://www.reddit.com/r/java/comments/ga1m4f/comment/fp1fsic/?utm_source=share&utm_medium=web2x&context=3

Here you have a situation of the team that's least familiar with OpenJDK, least engaged, and contributing the least also happens to be the only one doing something extremely misleading (the similarity of the OpenJ9 SDK and OpenJDK is less than 60%) and explicitly forbidden.

[koppor]

Current build blocked by https://bugs.openjdk.org/browse/JDK-8240567

[koppor]

OMG (1), they switched their distribution "name" (and kind) -- eclipse-openj9/openj9#15060

OMG (2), they do not have any JDK21 build released to SDKMan (yet).

[koppor]

Refs JabRef#11298

[koppor]

"semeru" is the packaging of OpenJ9 (see https://developer.ibm.com/languages/java/semeru-runtimes/downloads/). It is available as IBM at the toolchains - https://github.com/gradle/foojay-toolchains

[koppor]

Current build - on macOS:

Error: Hash of java.rmi (03f539ebf91c81cfc7a81236d2c1109e0d050c839d03c6f757022b99a10cb819) differs to expected hash (07126a5cefcc6c19d9059cfa4b42ec832f3714a1ef89f103afc2d1ab61b79bc7) recorded in java.base

Asked at https://openj9.slack.com/archives/C862YFGL9/p1716031780540649

[koppor]

Blocked by ibmruntimes/Semeru-Runtimes#3

[koppor]

We should not try the "balanced" GC first - "just" switch to OpenJ9.

Should be possible with JDK25. Blocked by JabRef#14057.

I will close - we need to create another PR with "just" IBM as vendor.