RFE: SERVER jvm into JRE

[javazoid]

Here’s the RFE:

http://developer.java.sun.com/developer/bugParade/bugs/4401674.html

Mik

[Bombadil]

“State: Closed, will not be fixed”

In case one day I’ll manage to finish a Java game project I’ll offer several versions of the game on my Web site: a JAR bundle, a Webstart version, and for the Windows users a special “JAR + JRE + Server VM’s dll + Starter.bat”. (The JRE without optional packages).

Jdk/Jre/Readme.txt:

---

Redistribution of Java 2 SDK Files

The limited set of files from the SDK listed below may be included in
vendor redistributions of the Java 2 Runtime Environment. All paths
are relative to the top-level directory of the SDK.
(…)

• jre\bin\server
  On Microsoft Windows platforms, the Java 2 SDK includes both the Java HotSpot Server VM and Java HotSpot Client VM. However, the Java 2 Runtime Environment for Microsoft Windows platforms includes only the Java HotSpot Client VM. Those wishing to use the Java HotSpot Server VM with the Java 2 Runtime Environment may copy the SDK’s jre\bin\server folder to a bin\server directory in the Java Runtime Environment. Software vendors may redistribute the Java HotSpot Server VM with their redistributions of the Java Runtime Environment.

Doing so with the international JRE1.4.2_04 version, I get:
Zip: 12,5 MB
Cab: 9,5 MB

[princec]

Ok, how about a new RFE: Combine server and client VMs into a single VM with two-stage compilation? It’s the Holy Grail Of Java Virtual Machine Implementation! At last, Java will live up to the performance claims that everyone makes for it*

Cas

*which are invariably made about the server VM, after it’s been running for 2 days continuously…

[javazoid]

Merging the two jvms is a nice suggestion, as the original SUN design of having multiple (pluggable ?) JVMs in the same runtime seems to be miserably failed.

But my question is: Is it really expensive to have the server jvm inside the JRE ? I estimated only 977Kb more, I think this extra overhead is acceptable in order to gain some extra speed for almost every java application.

Maybe the point is the quality of the client JVM: why we get such a different speed by running the same application with client and server jvms ? The server JVM outperforms the client one in math and array handling. So why do not include those code-optimizing parts of the server jvm into the client jvm by default ?
The server jvm could still be there for huge-memory-usage multithreaded server applications, while the client jvm would perform better for client applications such as games, gfx handling etc.

[princec]

I’m sure the server VM could be pretty trivially adapted to do quick and dirty client VM jitting after 1500 invocations and then discard it and perform the heavy server optimisations at 10000 invocations. How radically different can the two codebases be? Don’t answer I know the Sun engineers are thinking about it and have been for, ohh, two years now, but its mysteriously low on their list of priorities.

Cas

[genepi]

Yes!
I add my voice for a compiler dynamically optimizing with run time: http://forum.java.sun.com/thread.jsp?forum=316&thread=501335&tstart=0&trange=15