On 09-11-2012, at 7:22 AM, "David T. Lewis" <lewis at mail.msen.com> wrote:

> On Fri, Nov 09, 2012 at 01:43:15PM +0100, Bert Freudenberg wrote:
>> How long does a full GC take for, say, an 8 GB image with a gazillion of small objects?
>> 
>> - Bert -
> 
> Sorry Bert, a gazillion requires the 128-bit object format ;)
> 
> My totally unscientific observation is that growing the object memory by
> filling it with large objects has no noticeable impact on GC performance.
> I don't have any measurements to offer, but I think we will find that
> GC performance will degrade as a (non linear) function of number of
> objects, and size of the objects will have no impact.


Assuming not too much has been changed in the GC stuff, large objects tend to end up in the old-space region and don't get looked at very often. To torture the incremental gc code try something like creating a tree of associations, each with a key of another association and a value of another one. Repeat for perhaps 30 million. Go make dinner…

Oh, wait, Bert did specify a full GC; well that will of course scan all those big objects in old space too, so add 'grow your vegetables first'. Unless they are predominantly non-pointer objects, which get skipped over.

tim
--
tim Rowledge; tim at rowledge.org; http://www.rowledge.org/tim
Useful random insult:- Looks for the "Any" key.