From: Sylvain Joyeux <sylvain.joyeux@...4x.org>
Date: 2007-10-02T16:20:07+09:00
Subject: Re: gc.c -- possible logic error?

> Would you mind pointing me at your "object allocation
> tracking" patch"?
It is simply tracking the count of objects on the heap which are allocated. No 
tracking of freed objects. ruby-prof has a support for the patched ruby, so 
you get a profiling tool for object allocation (you want to check some 
patches I submitted to fix the latest ruby-prof as well though)

http://rubyforge.org/tracker/?func=detail&aid=11497&group_id=426&atid=1700

> Could you tell me more about your fixed-size heap
> allocation scheme?
Well, seeing you're at MBARI, I think we have the same kind of 
applications ... The main problem I am trying to address with the fixed-size 
allocation scheme is the GC time: I'm more concerned about latency than 
speed, and GC is really a great latency hog. I have no great results so far: 
it reduces the GC time 10% but GC is called more often. I did not have time 
to profile garbage_collect since I wrote the thing, so I cannot improve it 
for now.

The idea behind the fixed-size allocation scheme is to have small, fixed-size 
heaps (8192 objects for now). I was wondering wether or not it would gain 
because the probability of a freed heap would be greater. Since there is more 
heaps, I had to tweak the heap allocation so that is_pointer_to_heap remains 
fast: it is called often during GC...

Sylvain