Generational collection with mmap protection barriers is probably a good
strategy at this point.
Kevin
Donald Hunter wrote:
Allison,
I hacked out arena based GC and hacked in Boehm GC several months
back. IIRC for an execution run of compiled Rakudo code the
performance difference was minimal. Reading the IRClog it chimes with
what I remember - it would appear to be due to the fact that so many
short-lived objects are created, any GC is going to be under huge
pressure.
It will take me a while to resurrect my Boehm hack since I am in the
middle of a Windows 7 upgrade but I can gather some specific figures
if it will help.
Cheers,
Donald.
2009/11/24 Allison Randal <[email protected] <mailto:[email protected]>>
It's beginning to look like the next big Parrot task (along the
lines of the calling conventions refactor) will be the GC, as an
important performance enhancement for the compiler tools. See:
http://irclog.perlgeek.de/parrot/2009-11-20#i_1751189
Likely after 2.0. This is just a general message to get people
starting to think about it.
Allison
_______________________________________________
http://lists.parrot.org/mailman/listinfo/parrot-dev
------------------------------------------------------------------------
_______________________________________________
http://lists.parrot.org/mailman/listinfo/parrot-dev
_______________________________________________
http://lists.parrot.org/mailman/listinfo/parrot-dev
