Re: [Gc] RE: MPROTECT_VDB with PARALLEL_MARK

Ivan Maidanski ivmai at mail.ru
Wed Jan 12 14:43:29 PST 2011


Hi Damian and Hans,

I've just removed that undef MPROTECT_VDB caused by PARALLEL_MARK since I hadn't observed anything breaking the collection process.

So, generational collection is now enabled on multi-cores for most platforms.
Testing on various platforms (./configure --enable-parallel-mark ; make check) is appreciated.

Regards.

Wed, 12 Jan 2011 06:23:07 +0000 "Boehm, Hans" <hans.boehm at hp.com>:

> I'm not sure whether anything breaks completely.  I think it's designed not
> to.  However IIRC the combination of incremental GC and parallel marking:
> 
> a) Doesn't quite do what you expect, since GC_do_parallel_mark gets invoked
> even for an incremental GC and runs to completion, and
> b) May have inherent problems with the current algorithm since the incremental
> GC steps may be too small to warrant starting up the helper threads on each
> step.
> 
> Generally, I haven't had a chance to experiment enough with this.	  Probably
> the most useful and easiest configuration to try to get to work is pure
> generational mode, i.e. GC_incremental && GC_time_limit == GC_TIME_UNLIMITED.
> 
> Hans
> 
> > -----Original Message-----
> > From: Ivan Maidanski [mailto:ivmai at mail.ru]
> > Sent: Friday, January 07, 2011 2:38 AM
> > To: gc at linux.hpl.hp.com; Boehm, Hans
> > Subject: MPROTECT_VDB with PARALLEL_MARK
> > 
> > Hello Hans,
> > 
> > Could you explain me the reason of disabling MPROTECT_VDB if parallel
> > marking is on?
> > Thanks.



More information about the Gc mailing list