Re: [rvm-research] Making Stop-the-world on-the-fly

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

I think that for what you're doing, you'll have to do a lot of work in  
the MMTk<->RVM glue as well as possibly the RVM itself.  That's my  
experience with on-the-fly collection anyway.  The thread stop code is  
in the RVM side of the MMTk<->RVM glue.  Here's a rough picture of how  
the system works:

RVM <---> RVM side glue to MMTk <---> MMTk side glue to RVM <---> MMTk

The RVM code is in rvm/src.  The RVM side of the glue is in rvm/src/ 
org/jikesrvm/mm/mminterface.  The MMTk side glue to RVM is in MMTk/ext/ 
vm/jikesrvm/org/jikesrvm/mm/mmtk.  MMTk itself is in MMTk/src.

The functionality is split roughly as follows:

RVM: VM functionality not including memory management, such as  
compilation, class loading, locking, native calls, threads, etc.

MMTk: memory management implementation that abstracts over anything VM- 
specific; i.e. it doesn't even really have a notion of "thread" only  
that there are "mutator contexts" that may or may not have some state  
attached to them, and that state may or may not be in the form of a  
stack.  MMTk doesn't know how to start/stop threads, and really  
doesn't even know that there is even a need to do so, other than  
having some vague distinctions between a StopTheWorld collector and a  
concurrent one.

RVM side glue to MMTk: this is the code that decides to invoke a  
collection, stops the world, and transfers control to MMTk.  I think  
it also contains some object and stack scanning logic.

MMTk side glue to RVM: this code gives MMTk access to the VM.  for  
example, if MMTk wants to do some locking, the MMTk side glue to RVM  
will give it a lock that is implemented using RVM's functionality.   
There are other things in here, like accessing the RVM object models,  
etc.

This, if you want to understand how threads are started and stopped,  
you want to look at the RVM side of the glue, in rvm/src/org/jikesrvm/ 
mm/mminterface.  That will have calls to RVM-internal code such as rvm/ 
src/org/jikesrvm/scheduler/RVMThread.java - where the actual thread  
control logic resides.  On the RVM side we've already implemented a  
lot of what is needed to have a sliding views collector - in  
particular the RVMThread API already has a notion of a soft handshake  
that can be configured (by hacking the code ;-)) to do pretty much  
anything you want, and that may include stack scanning, GC context  
flushing, or even just a no-op to get an acknowledgement of GC phase  
change.

I would just like to add that I concur with Eliot - you're jumping  
into the RVM with the intent of solving a problem that people with  
extreme familiarity with the RVM would have trouble doing.  I've  
implemented concurrent, on-the-fly collectors, and have done a lot of  
RVM hacking, but I'd still categorize what you're doing as difficult  
for me.  That said, ambition is a good thing. :-)

So, good luck,
-Filip

On Oct 14, 2009, at 2:58 PM, Lalit Sharma wrote:

> Hi,
> Thanks Mike for such a quick reply. I'm using eclipse, I've followed  
> the tutorials to convert the nogc to marksweep and so on. But, i'm  
> not able to understand where are the methods which stop a perticular  
> thread.
> I just wanted to do some benchmarks.
> ------------------------------
>
> Message: 4
> Date: Wed, 14 Oct 2009 23:03:51 +0530
> From: Lalit Sharma <lal...@gm...>
> Subject: [rvm-research] Making Stop-the-world on-the-fly
> To: jik...@li...
> Message-ID:
>        <5f7...@ma...>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi,
> I am trying to convert the stoptheworld collector to an on the fly  
> one,
> using the sliding views concept. I'm finding it very hard to even  
> understand
> the jikes RVM code. The papers I'm referring to have already  
> implemented the
> code.
> Please help me understand what changes I have to do.
> I am completely confused.
>
> --
> N. Lalit Vidyadhar Sharma
> lal...@gm...
> -------------- next part --------------
> An HTML attachment was scrubbed...
>
> ------------------------------
>
> Message: 5
> Date: Wed, 14 Oct 2009 13:43:13 -0500 (CDT)
> From: Michael Bond <mik...@cs...>
> Subject: Re: [rvm-research] Making Stop-the-world on-the-fly
> To: "General discussion of Jikes RVM design, implementation, issues,
>        and plans" <jik...@li...>
> Message-ID: <alp...@gr...>
> Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed
>
> Hi Lalit,
>
> First, please do edit in Eclipse or some other IDE
> (http://jikesrvm.org/Editing+JikesRVM+in+an+IDE) -- it'll really  
> help with
> understanding the code.  It sounds like you'll mainly be working in  
> MMTk
> (the Memory Management Toolkit), i.e., the GC component of Jikes  
> RVM, and
> it's highly object oriented, so you really need Eclipse.
>
> Second, fortunately MMTk has nice documentation:
>
> - MMTK user guide: http://jikesrvm.org/MMTk
>
> - tutorial for getting started building new collectors:
> http://jikesrvm.org/MMTk+Tutorial
>
> Beyond that, it does take a while to be able to understand MMTk or  
> other
> parts of Jikes.  Copious browsing of revelent code is perhaps the best
> approach :)
>
> Cheers,
> Mike
>
>
> On Wed, 14 Oct 2009, Lalit Sharma wrote:
>
> > Hi,
> > I am trying to convert the stoptheworld collector to an on the fly  
> one, using the sliding views concept.
> > I'm finding it very hard to even understand the jikes RVM code.  
> The papers I'm referring to have already
> > implemented the code.
> > Please help me understand what changes I have to do.
> > I am completely confused.
> >
> > --
> > N. Lalit Vidyadhar Sharma
> > lal...@gm...
> >
> >
>
>
>
> ------------------------------
>
> ------------------------------------------------------------------------------
> Come build with us! The BlackBerry(R) Developer Conference in SF, CA
> is the only developer event you need to attend this year. Jumpstart  
> your
> developing skills, take BlackBerry mobile applications to market and  
> stay
> ahead of the curve. Join us from November 9 - 12, 2009. Register now!
> http://p.sf.net/sfu/devconference
>
> ------------------------------
>
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jik...@li...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>
>
> End of Jikesrvm-researchers Digest, Vol 41, Issue 3
> ***************************************************
>
>
>
> -- 
> N. Lalit Vidyadhar Sharma
> aka: 'Lalli'
> lalit.shorl.in
> lal...@gm...
> ------------------------------------------------------------------------------
> Come build with us! The BlackBerry(R) Developer Conference in SF, CA
> is the only developer event you need to attend this year. Jumpstart  
> your
> developing skills, take BlackBerry mobile applications to market and  
> stay
> ahead of the curve. Join us from November 9 - 12, 2009. Register now!
> http://p.sf.net/sfu/devconference_______________________________________________
> Jikesrvm-researchers mailing list
> Jik...@li...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers