Re: Tunable overcommitting
Snow Cat (snowcat@netgate.net)
Mon, 7 Apr 1997 09:49:22 -0700 (PDT)
Mikko J Rauhala once wrote:
>
> Greetings
>
> As we all know, there are two ways to get something implemented in the
> Linux kernel: Do it yourself or bitch about it on the kernel mailing
> list and hope that somebody does it. Here I am, doing the latter:
>
> Memory overcommitting is a relatively good idea in the world of
> memory-hogging software. However, sometimes the results are not too nice.
> Now, I was wondering that it'd be nice to have the maximum amount be
> tunable, for example via a /proc entry. The amount of overcommitting
> allowed could be presented in pages, so that the tuning is as
> fine-grained as possible throughout the memory area (so that you could
> disable overcommitting altogether or not to impose any limits). If the
> parameter would be changed and the overcommitting would already be
> higher than that, then all requests for more memory would fail until
> enough would be freed.
>
> Anyway, just a suggestion. Maybe, if nobody else will do it, I will look
> into it more someday when I have the time. Should learn some kernel hacking
> anyway...
>
Actually, memory hogs can control their own overcomittment by using
mmap() to allocate memory and passing MAP_NORESERVE to the pages that
they don't want checked.
--
_. _ .
(_ ,_ _ , . / ` _ _L | Email: Oleg Kibirev <snowcat@netgate.net>
._)| U(_)\/\/ \_,(_L/L | http://ng.netgate.net/~snowcat/
------------------------'