[kaffe] native threads

Mihai Surdeanu mihai at seas.smu.edu
Fri Feb 12 14:20:31 PST 1999


> >The port by itself wouldn't be.  Optimizing a VM for an SMP might be
> >a thesis paper.
This is what I'm working on for mine... I see some nice issues in it:
processor allocation algorithms, memory consistency models and distributed
garbage collection. I believe that these issues have not been really
discussed from the Java perspective (especially for systems without shared
memory).

> >Hard to say how long it would take: it depends on how familiar you
> >are with multi-threaded programming and how familiar you are with Kaffe.
> >Could take anywhere from a day to two weeks.

For what it's worth:
I started to work in Dec 98 on a distributed VM (based on kaffe) for a
network of workstations. I had no idea what's inside kaffe. It took me
about 2 months :) to have a basic system (no gc, no awt) running (almost
stable). I'm using jthreads and jit on linux. The system transparently
distributes threads, objects and locks over the network. Now I have to
start worrying about the real stuff: Java oriented processor allocation
algorithms, object consistency protocols (single-writer vs multiple
writers, lazy release vs eager release vs ...) etc.

Mihai Surdeanu		Southern Methodist University, CSE 
			(214) 768 - 3054



More information about the kaffe mailing list