[gclist] Program behaviour

[Nick Barnes]

> >This surprises me somewhat, given that free objects on a page should be
> >allocated to consecutive requests.  But distance is not a good measure of
> >locality.  How many pages are spanned by those cycles?
> 
> This is a very good question.
> 
> In general, distance in address space doesn't matter at all as long as
> things that are touched at about the same time are clustered onto a few 
> pages.
> 
> This isn't strictly true, because you may get locality effects at a
> larger-than-page granularity because of disk geometries and caches
> in the disk drive, e.g., accessing several blocks in one track
> may win relative to accessing several blocks scattered over several
> tracks.

Also TLB entries, &c.

Nick Barnes