New subject: [xwiki-devs] Does JDK6u14 "Garbage-First garbage collector (G1)" work and/or improve Xwiki performance/size/mem-locality?

18 Feb 2009

I usually have found that the garbage collector is the hardest-working thing
in a java-web app (unfortunately) so when I noticed this new GC option in *Java
HotSpot 14*, I figured it might help (potentially a lot w/ high-volume
sites, which are the real GC churners).

Has anybody tried the "Garbage-First garbage collector (G1)"  with Xwiki?

http://platform.xwiki.org/xwiki/bin/view/AdminGuide/Performances suggests
...

 CATALINA_OPTS="-server -Xms128m -Xmx1024m -XX:MaxPermSize=128m -Dfile.encoding=utf-8
-Djava.awt.headless=true -XX:+UseParallelGC -XX:MaxGCPauseMillis=100"

 However these instructions don't work, as-is, since UseParallelGC is the default
nowadays: 
https://jdk6.dev.java.net/6uNea.html says:

The parallel collector is still the default GC and is the most efficient GC
...
  for common household usage. G1 is meant to be an
alternative for the
 concurrent collector. It is designed to be more predictable and enable fast
 allocation with memory regions design. 

Having actually worked on a variety of strange garbage collectors (for Lisp)
where all of java's problems had already been solved decades ago ...this
feels like a 1985-1990 flashback. :-)

Here's more info

Java SE 6 Update 14 Milestone Schedule b01 (1st available build) 02/11/09  FCS
...
  Q2, 2009 *Feature List:*

    - *Service Tag creation for Windows JRE (CR
6676849<http://bugs.sun.com/view_bug.do?bug_id=6676840>
    )*: For more information about Service Tag, check out this technical
    article
    <http://java.sun.com/developer/technicalArticles/productregistration/>.

    - *Java HotSpot 14*: including the new garbage collector, G1, Big
    Decimal enhancements, TreeMap and HashMap enhancements, optimized compressed
    OOP code generation.

 The Garbage-First garbage collector (G1) is currently in beta. It is not
 enabled by default. The parallel collector is still the default GC and is
 the most efficient GC for common household usage. G1 is meant to be an
 alternative for the concurrent collector. It is designed to be more
 predictable and enable fast allocation with memory regions design.

 To use it: -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC

 For more information about GC & G1, please see:

    - G1 Technical Session in JavaOne
2008<http://developers.sun.com/learning/javaoneonline/j1sessn.jsp?sessn=…
    - Java SE 6 GC
tunning<http://java.sun.com/javase/technologies/hotspot/gc/gc_tuning_6.h…
    - Analysis of the Garbage-First Garbage
Collection<http://research.sun.com/jtech/pubs/04-g1-paper-ismm.pdf>

 Niels
http://nielsmayer.com

PS: and will a different GC prevent sequences like this from happening?? see
http://nielsmayer.com/happy-happy-joy-joy.txt

PPS:  some things that cause big-growth, like importing and exporting, might
not grow as large with a "generation scavenging" style GC as provided by
"Garbage-First Collection." Sometimes, just GCing a large structure being
iterated-over uses a lot more memory than it needs to; because the gc is
letting objects that should get collected early, like the incremental
results of an iteration, "build up"  and increase overall memory size while
decreasing locality and cache-hits. This seems to cause a nearly exponential
performance dropoff when very little memory churn might occur if only things
got collected "at the right time."

http://research.sun.com/jtech/pubs/04-g1-paper-ismm.pdf suggests this new GC
will help:

2.4 Generational GarbageFirst
...

 Generational garbage collection [34, 26] has several advantages, which a
 collection strategy ignores at its peril. Newly allocated objects are
 usually more likely to become garbage than older objects, and newly
 allocated objects are also more likely to be the target of pointer
 modications, if only because of initialization. We can take advantage of
 both of these properties in Garbage-First in a flexible way. We can
 heuristically designate a region as young when it is chosen as a mutator
 allocation region. This commits the region to be a member of the next
 collection set. In return for this loss of heuristic flexibility, we gain an
 important benefit: remembered set processing is not required to consider
 modifications in young regions. Reachable young objects will be scanned
 after they are evacuated as a normal part of the next evacuation pause.

 Note that a collection set can contain a mix of young and non-young
 regions. Other than the special treatment for remembered sets described
 above, both kinds of regions are treated uniformly.

 ...

 2.5 Concurrent Marking

 Concurrent marking is an important component of the system. It provides
 collector completeness without imposing any order on region choice for
 collection sets (as, for example, the Train algorithm of Hudson and Moss
 [22] does). Further, it provides the live data information that allows
 regions to be collected in \garbage-first" order. This section describes our
 concurrent marking algorithm.