> > I'd like to point out that, assuming what people have told me is true,
> > you don't get 12MB of virtual memory out of that setup.  Apart from
> > the fact that the kernel and any resident text pages come out of the
> > 4MB of physical memory, once a swap page is allocated for a given vm
> > object, it's never deallocated until the object goes away.

[ ... ]

> re: the objects not getting collapsed: "that's a bug, and it should be
> fixed."  Additionally, it's not clear to me how much that matters on
> 4M machines (sure, it's a problem because of less ram, but people
> _really_ shouldn't be putting heavy load on them).

If I understand David correctly, he wasn't speaking about the non-collapsing
bug, but was talking about the fact that swap space allocated to back some
area of memory (e.g. anonymous memory, or shadow copies of some other area)
will not be freed even when all the pages of this swap chunk are brought back
into memory. It will remain allocated as long as the anonymous memory remains
in use, that is, in most circumstances until the concerned process dies.

AFAIK, there is no intent by anyone to change this.
--
ws@TooLs.DE     (Wolfgang Solfrank, TooLs GmbH) 	+49-228-985800