Re: [rvm-research] Modifying the JVM to have a more deterministic garbage collection model

[<mi...@gm...>]

Hello All,
 I am working on my predictive garbage collection project using reference
counting. I looked at MMTK source code and I can see that there already
exists a refrence counting GC implementation. I compiled the rvm using the
RefCount config and ran java programs on it.Every thing seems to be working,
except what I have noticed is that finalize never gets called when I am
running on rvm where as when I run the same code over the jvm all objects
get finalized. Is this by design?

Thank you

On Mon, Nov 10, 2008 at 1:28 AM, <mi...@gm...> wrote:

> Hi,
> For my class project I want to modify the JVM to have a more deterministic
> garbage collection model. I will take a  reference counting approach to
> accomplish this.  More specifically I will use a variation of  the algorithm
> described here
>
>
> http://www.jucs.org/jucs_13_6/cyclic_reference_counting_with/jucs_13_6_0830_0838_lins.pdf
>
> In order to do this I need to store  reference count or multiple reference
> counts for each object.  I will also need to update these counts when a
> reference is added or removed. I need some help on figuring out what would
> be the best way to do this? Can this be done by tweaking or replacing some
> parts of the jvm or would I have to do an over haul of the whole jvm (I hope
> that is not the case :) ) I also wanted to know where I would put hooks to
> update refrence counts and what would be the best place to store the
> reference counts, with in the object itself or somewhere else. Any
> suggestion on software engineering aspect of this project would also be very
> helpful.  I wanted to compile and run the jvm on windows2000 and from the
> little research I have done seems like windows implementation of jikes is
> still in works and wanted to know how far along was it.
>
>
> Thank you