Re: process creation time increases linearly with shmem

[Ross Biro <ross.biro@gmail.com>]

On 8/26/05, Hugh Dickins <hugh@veritas.com> wrote:
> On Fri, 26 Aug 2005, Ross Biro wrote:
> > On 8/26/05, Hugh Dickins <hugh@veritas.com> wrote:
> > >
> > > The refaulting will hurt the performance of something: let's
> > > just hope that something doesn't turn out to be a show-stopper.
> >
> > Why not just fault in all the pages on the first fault. Then the performance
> > loss is a single page fault (the page table copy that would have happened a
> > fork time now happens at fault time) and you get the big win for processes
> > that do fork/exec.
> 
> "all" might be very many more pages than were ever mapped in the parent,
> and not be a win.  Some faultahead might work better.  Might, might, ...

If you reduce "all" to whatever would have been done in fork
originially, then you've got a big win in some cases and a minimal
loss in others, and it's easy to argue you've got something better.

Now changng "all" to something even less might be an even bigger win,
but that requires a lot of benchmarking to justify.

    Ross