JVM Ergonomics

[Krystal Mo]

Hi Joseph,

Just to point you to the source code where argument parsing and a part 
of the ergonomics are:
http://hg.openjdk.java.net/jdk7u/jdk7u/hotspot/file/db63a909e1ad/src/share/vm/runtime/arguments.cpp

line 2936: jint Arguments::parse(const JavaVMInitArgs* args)
This is the main entry point of argument parsing.
Within that, the call to parse_vm_init_args(args) at line 3029 is the 
place where most of the command line arguments are parsed.
After that, the calls to set_*() in Arguments::parse() is where most 
ergonomic decisions are made; not just the call to set_ergonomics_flags().
Some of the ergonomic logic take place only if the corresponding VM flag 
is not explicitly set; some may set a couple of flags depending on the 
value of another flag (take -XX:AggressiveOpts for example).

Hope it helps,
Kris

On 08/10/2012 03:03 AM, Joseph D Carroll Jr wrote:
> I am working on a blog post about the JVM ergonomics and I am a little
> stuck.  The blog post is related to Eclipse and the runtime arguments it
> uses.  I am looking for the point at which the detected ergonomic settings
> are merged with the command line arguments, does the VM parse the command
> line args first or determine the ergonomics first (when all of that
> occurs).  I am also looking for how the JVM detects what ergonomic settings
> it should use, I realize that this will be platform specific.  However, I
> cannot seem to find any technical documentation about these topics beyond
> that of a general ergonomics overview and after reviewing the source of
> openJDK I haven't been able to determine anything.
>
> Would someone have any insight into this or know who I would be able to
> speak with that could point me in the right direction?
>
> My objective is to create(/propose) a process where you are able to
> configure the Eclipse launchers with something like
> --launcher.MaxHeapNotLessThan=512M in the Eclipse launcher configuration.
>   This would result in a platform specific call to determine the underlying
> hardware configuration much in the same way the JVM does.  Then, a number
> of simple checks could be performed and the JVM could be launched with the
> optimal arguments.
>
> Eclipse presently defines a number of vm args that, in my opinion, are
> sub-optimal.  I also do not feel that many people are aware of the
> ergonomic functionality.  So the blog post is going to describe ergonomics,
> give a little more detail than normal with some performance comparisons,
> and then I was going to propose the above change.
>
> Any comments / questions / or anything else is certainly welcome.
>
> Thanks,
>
> JD