Re: Note describing poor dcache utilization under high memory pressure

[Daniel Phillips]

On January 30, 2002 04:54 pm, Rik van Riel wrote:
> On Wed, 30 Jan 2002, Daniel Phillips wrote:
> > On January 30, 2002 03:46 pm, Rik van Riel wrote:
> > > On Wed, 30 Jan 2002, Daniel Phillips wrote:
> 
> > > >      |-bash---bash---xinit-+-XFree86
> > > >      |                     `-xfwm-+-xfce---gnome-terminal-+-bash---pstree
> > >
> > > It doesn't matter how deep the tree is, on exec() all
> > > previously shared page tables will be blown away.
> > >
> > > In this part of the tree, I see exactly 2 processes
> > > which could be sharing page tables (the two bash
> > > processes).
> >
> > Sure, your point is that there is no problem and the speed of rmap on
> > fork is not something to worry about?
> 
> No.  The point is that we should optimise for fork()+exec(),
> not for a long series of consecutive fork()s all sharing the
> same page tables.

Fork+exec is adequately optimized for.  Fork+100 execs is supremely well
optimized for.  I'm entirely satisfied with the way the performance looks
at this point, it will outdo anything we've seen to date.  With Linus's
write-protect-in-page-directory optimization there's not a lot more fat
to be squeezed out, if any, and even without it, it will be a screamer.
I think we've done this one, it's time to move on from here.

-- 
Daniel