Re: New generational GC Scheme

2005-06-17 Thread Leopold Toetsch
Alexandre Buisse wrote: On 6/8/05, Leopold Toetsch <[EMAIL PROTECTED]> wrote: 1) a local GC run in the threads arena 2) GC of the shared arena Case 1) is simple as it's fully private to the thread. A non-shared object can not be referenced by a shared one, therefore all local garbage can be i

Re: New generational GC Scheme

2005-06-16 Thread Alexandre Buisse
Sorry for answering so late... On 6/8/05, Leopold Toetsch <[EMAIL PROTECTED]> wrote: > jerry gay wrote: > > i'm no gc expert, but here's my comments after discussions with > > alexandre on #parrot. > > > > On 6/8/05, Alexandre Buisse <[EMAIL PROTECTED]> wrote: > > > since threading issues haven't

Re: New generational GC Scheme

2005-06-08 Thread Leopold Toetsch
Sam Vilain wrote: AIUI, another effect of adding another indirection to every pointer lookup, is that it will place a little extra stress on the memory caching systems, as two regions will need to be accessed continually. Yes, that's of course true. But please compare: now

Re: New generational GC Scheme

2005-06-08 Thread Sam Vilain
Alexandre Buisse wrote: So we add a level of indirection: objects consist of a header and the actual data. Headers are allocated once and for all at object creation and do not move. [...] The big disadvantage of this approach is that we use one or two words (if objects need to know where their he

Re: New generational GC Scheme

2005-06-08 Thread Leopold Toetsch
jerry gay wrote: i'm no gc expert, but here's my comments after discussions with alexandre on #parrot. On 6/8/05, Alexandre Buisse <[EMAIL PROTECTED]> wrote: since threading issues haven't been mentioned here, i thought i'd make sure they get discussed. now, if someone could say something int

Re: New generational GC Scheme

2005-06-08 Thread Klaas-Jan Stol
"tuned dynamically", as in at run-time? alexandre mentioned this may be possible on #parrot, but there may be trouble with decreasing the generation count. i don't know if run-time tuning of the generation count is necessary, but if so, i imagine that decreasing the count at run-time could be ac

Re: New generational GC Scheme

2005-06-08 Thread jerry gay
i'm no gc expert, but here's my comments after discussions with alexandre on #parrot. On 6/8/05, Alexandre Buisse <[EMAIL PROTECTED]> wrote: > Following google SoC and TODO item, > https://rt.perl.org/rt3//Ticket/Display.html?id=33922, here is the > scheme proposal for a new parrot GC (many thanks