Re: [PATCH] VM strict overcommit

[Alan Cox]

On Sun, 2002-07-21 at 00:57, Robert Love wrote:
> No problem.
> 
> > Lets go with a sysctl tuned value and see what the 2.5 world finds the
> > best numbers to be ?
> 
> Great idea.  This allows pedants with swap to use 0, others to use 50,
> and those without swap to pick whatever works for them (e.g. 65% as you

On a swapless box you are unlikely to ever achieve 65% used by anonymous
pages in real world applications. Its suprising how much of memory is
file backed when you actually try it out. One exception in the future
might be a system with extensive amounts of XIP, where very little of
the binaries is actually pulled into RAM and thus counted into the
figures.