[
https://issues.apache.org/jira/browse/FLINK-2343?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14623832#comment-14623832
]
Yanping Wang commented on FLINK-2343:
-------------------------------------
Hi, Stephan
I'd like to promote G1 as default GC for Flink with following reasons
(1) CMS is no longer the development focus from Oracle Java and OpenJDK, all
the GC development is focusing on G1
(2) Based on jdk9 plan, jdk9's default collector for jdk9 release will be G1
(3) jdk8 update 45 is now the default jdk to be downloaded from
http://www.oracle.com/technetwork/java/javase/downloads/index.html In jdk8
update 45, G1 is in full production with many optimizations fit for Spark, etc
running 100GB plus heap.
(4) In addition, upcoming release of jdk8 update 60 will contain following
optimizations on:
• Refactor Remembered Set for Object Array
• Early reclaim dead humongous objects
• Hot card cache lock contention reduction
• ext root scan optimization (scanning the roots: globals, registers,
thread stacks and VM data structures)
• class unloading during the marking cycle
• object copy optimization
• G1 default value adjustment for some global flags
• Dynamic IHOP
• Fix PLAB sizing heuristic
I used jdk8 update 60 (downloaded from here:
https://jdk8.java.net/download.html ) measured 2X performance improvement
over jdk7 update 79 on one Spark graph computing workload running 4 node
clusters of Intel Ivybridge servers with 160GB java heap.
I was the one of several JVM runtime engineers who helped putting those
optimizations for Spark in jdk8 update 40 and above. Let me know if you have
any questions regarding G1 status and how to best run G1.
> Change default garbage collector in streaming environments
> ----------------------------------------------------------
>
> Key: FLINK-2343
> URL: https://issues.apache.org/jira/browse/FLINK-2343
> Project: Flink
> Issue Type: Improvement
> Components: Start-Stop Scripts
> Affects Versions: 0.10
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
> Fix For: 0.10
>
>
> When starting Flink, we don't pass any particular GC related JVM flags to the
> system. That means, it uses the default garbage collectors, which are the
> bulk parallel GCs for both old gen and new gen.
> For streaming applications, this results in vastly fluctuating latencies.
> Latencies are much more constant with either the {{CMS}} or {{G1}} GC.
> I propose to make the CMS the default GC for streaming setups.
> G1 may become the GC of choice in the future, but fro various articles I
> found, it is still somewhat in "beta" status (see for example here:
> http://jaxenter.com/kirk-pepperdine-on-the-g1-for-java-9-118190.html )
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
