Re: The hall of bloat

[Behdad Esfahbod <behdad cs toronto edu>]

On Mon, 16 May 2005, Soeren Sandmann wrote:

> > There is a cost if the mapping has just a few random pages filled as you
> > have to fill in the whole page table heirarchy down to the page in
> > question.
>
> Isn't it always the case that you have to fill in the page table
> hierarchy down to the pages that you are actually using, regardless of
> the size of the mapping? Why would a mapping of 8K with both pages
> used require more page table entries than a mapping of 10M with two
> pages used?

I'm no export here, but I think this is what Allan was saying:
If you have a 10M stack that only uses the last two couple of
pages, then the "page table" for the whole 10M will be
initialized, most of them not mapped though.  So the penalty is
allocating and initializing 10M/4k=2560 page structs.

> Søren

--behdad
http://behdad.org/