Re: [JOS-Kernel] New paper available on KaffeOS (a Java OS)

[Gilbert Carl Herschberger II]

At 04:31 AM 6/2/00 MDT, Jay Lepreau <[EMAIL PROTECTED]> wrote:
>Hi JOS'ers,

Thanks for the update. This is welcome news.

>You should be able to get some design ideas from this paper.

Yes, we should. I am pleased to here that Kaffe has been extended to build
KaffeOS. It seems to me that Kore is compatible with Sun's JDK 1.02. What
is Kaffe compatible with? I'd like to think KaffeOS can benefit from our
research, too.

>The KaffeOS architecture supports the OS abstraction of a 
>process in a Java virtual machine. Each process executes as if it
>were run in its own virtual machine, including separate garbage
>collection of its own heap.

It sounds like KaffeOS is similar to JNI design, where each bytecode
process has four threads: main, gui, gc and finalization. In MPCL design,
each bytecode process has three threads: main, gui and finalization. The gc
thread is vm-wide.

>Overall, KaffeOS is up to 11% slower than the JVM on which it is based,
>which is an acceptable penalty for the safety that it provides. KaffeOS is
>substantially slower than commercial JVMs, but exhibits much better
>performance scaling in the presence of uncooperative code.

Here is a comment worthy of more discussion. We expect our MPCL-compatible
virtual machine to be as fast (or faster) than decaf in host mode. We do
not plan on reworking every opcode method to deal with multiple bytecode
processes. As far as the interpreter (thread) is concerned, it is still a
virtual machine with a single primordial class loader.

>Drawing the Red Line in Java

This article sounds familiar. For more information, see also the
AboveAndBelow article on JOS Wiki.


_______________________________________________
Kernel maillist  -  [EMAIL PROTECTED]
http://jos.org/mailman/listinfo/kernel