At 05:38 PM 9/18/00 +0100, you wrote:
>Sorry to be negative, but I would say we'd have to pull something pretty
>amazing out of the hat, JVM-wise, to be able to beat HotSpot or IBM's JVM on
>Linux in any area at all. Personally, I think the only reason to be
>interested in writing a new JVM is for personal understanding of how they
>work and the intellectual challenge (at least that's why I've been
>involved).
I definitely can't argue this possibility. And I agree that we almost
assuredly won't be able to outcompete head to heard replicating a standard
os with a jvm on it. The hope is using a completely different paradigm
will give us a big boost in some way. I personally like the idea of being
a jini os, with a hyper optimized javaspace but who's to say. I agree that
jos that looks like java on linux will probably not be that great. But
something that doesn't looks like unix at all may be the ticket.
>This overlaps with what I was going to suggest as follow-up to Robert's 'The
>JOS Project?' thread. Perhaps JOS should develop along two lines:
>
>1. JVM / low-level OS development along the current lines. If this ever bore
>fruit, then it could replace Linux in the following:
>
>2. A separate project to work on closer integration of existing JVM's into
>Linux so that shell, user-interface etc is all written in Java. Kind of a
>Java-centric KDE / Gnome initially, but it could then worm it's way further
>down to start taking over file system and other duties, if we wanted. This
>gives us something of a Java OS but without the overhead of trying to take
>on the work of writing both Linux and a JDK from the ground up ... (NB
>something like this was suggested over at the JavaLobby .. anyone know what
>happened to it?)
This is exactly what most of the core developers have suggested. And Ryan
has been pushing, and Robert has been really pushing. For instance, even
without our own os and jvm, you can actually build almost the entire jos on
top of any jvm on any os. Like Ryan's rheise.os. It runs on a jvm. You
can do a heck of a lot of coding and testing without any special
javaos. This is even more evident if you bring in the concept of providing
authentication, GUI/desktop/window manager, applications etc. None, or
very few really require the low level jvm/os to code. And all could be
very useful without it.
Robert has been trying to bridge the completely pure java stuff, with the
kernel by creating a kernel interface. That way even code that has to do
real level stuff can actually be written and tested on any os/jvm combo by
essentially writing a jos-kernel driver that implements his kernel
interface. Check out rjk (an old copy should be lying around somewher).
Perhaps we should really stress these alternatives as the discussions
evidently never make it out of a limited group of kernel developers. Maybe
we should have codenames for them to really distinguish the choices and
emphasize the differences.
-iain
_______________________________________________
Kernel maillist - [EMAIL PROTECTED]
http://jos.org/mailman/listinfo/kernel
