Curious Kaffe vs. jdk speed test results under Linux
Godmar Back
gback at cs.utah.edu
Sun Jan 3 16:06:57 PST 1999
>
> On Sun, 03 Jan 1999, Godmar Back wrote:
> >..However, new native code (such as the awt
> >code) should be written to use JNI, and JNI versions of some of the native
> >classes are welcome too.
..
>
> The AWT already is JNIed. But you have to care for your design a lot if it comes
> to a complete move to JNI (speed-wise, "unhand()" is dramatically
> better than a callback). There is a talk I gave about this topic on the papers
> webpage ("going native with Javas JNI..", I think).
>
My take on this is that the complete move the JNI only makes sense if
accompanied by a move to a more advanced generational collector that
actually requires JNI and compensates for it.
Nevertheless, I am in favor of a transition path that converts all those
methods that are likely to not be frequently used even before such a gc is
available.
kaffe.lang.UNIXProcess.forkAndExec is one example. Clearly, what methods
are frequently used does depend on the application you're running.
There is two more options:
One other option that we have is to make the use of JNI optional; i.e.,
to maintain two versions of certain native libs, a JNI and a non-JNI version.
Even with a more advanced gc, we would also have a third option: namely
to handcraft the non-JNI library code such that it coexists with the
collector.
- Godmar
More information about the kaffe
mailing list