Note .. As an optimization, BG-RC eliminates the header in the nursery, and adds it only when performing integration of nursery objects into the RCspace . Interesting that removes the header create cost on new objects ( and many small objects dont last long ) . God knows how you do get runtime type ( though you could do it like C++ and put it in the virt table and for concrete objectes just replaces with static type.)
On Thu, Oct 17, 2013 at 11:56 PM, Bennie Kloosteman <[email protected]>wrote: > Speaking of fragmentation this is an interesting and modern paper im > wading through ( most of you probably know it though ) > > Filip Pizlo FRAGMENTATION TOLERANT REAL TIME GARBAGE COLLECTION ( Draft) > http://www.filpizlo.com/papers/pizlo-phd.pdf > > He also has 2 other papers not on fragmentation but are of interest to > real time GC ( didnt read past the first paragraph) > http://www.filpizlo.com/papers/kalibera-rtss2009-minuteman.pdf Scheduling > Hard Real-time Garbage Collection > http://www.cs.technion.ac.il/~erez/Papers/*stopless*.pdf Stopless: A > real-time garbage collector for > multiprocessors<http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.108.322> > . > > > On Wed, Oct 16, 2013 at 2:17 AM, Jonathan S. Shapiro <[email protected]>wrote: > >> Thanks, Sandro. Good reference. It was in my queue, but I hadn't gotten >> to it yet. >> >> >> On Tue, Oct 15, 2013 at 9:50 AM, Sandro Magi <[email protected]>wrote: >> >>> Ha, sorry about that, just Bacon et al's past papers on cycle collection: >>> >>> [1] >>> researcher.ibm.com/files/us-**bacon/Paz05Efficient.pdf<http://researcher.ibm.com/files/us-bacon/Paz05Efficient.pdf> >>> >>> >>> >>> On 15/10/2013 11:47 AM, Jonathan S. Shapiro wrote: >>> >>>> Misplaced referent. What is [1]. ?? >>>> >>>> On Tuesday, October 15, 2013, Sandro Magi wrote: >>>> >>>> On 15/10/2013 11:07 AM, Jonathan S. Shapiro wrote: >>>>> >>>>> If we want to have a sensible performance story on*any* of these >>>>>> machines, >>>>>> we really need to push hard to get the required heap footprint as >>>>>> close to >>>>>> (1 * actual need) as we can. >>>>>> >>>>>> >>>>> According to [1], RC is much better than tracing for tight heaps. Of >>>>> course, that paper doesn't really account for long-term fragmentation. >>>>> >>>>> Sandro >>>>> >>>>> >>>>> >>>> >>>> >>>> ______________________________**_________________ >>>> bitc-dev mailing list >>>> [email protected] >>>> http://www.coyotos.org/**mailman/listinfo/bitc-dev<http://www.coyotos.org/mailman/listinfo/bitc-dev> >>>> >>>> >>> >>> >>> _______________________________________________ >>> bitc-dev mailing list >>> [email protected] >>> http://www.coyotos.org/mailman/listinfo/bitc-dev >>> >>> >> >> _______________________________________________ >> bitc-dev mailing list >> [email protected] >> http://www.coyotos.org/mailman/listinfo/bitc-dev >> >> >
_______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
