RPI3 swap experiments (grace under pressure)

[Konstantin Belousov]

On Wed, Aug 15, 2018 at 05:55:13AM -0700, John-Mark Gurney wrote:
> Jedi Tek'Unum wrote this message on Tue, Aug 14, 2018 at 17:17 -0500:
> > I have one question??? when was the last time anyone saw Solaris kill a process because the system was under memory stress? In my experience, NEVER! And I wouldn???t say that the system became unreasonably unresponsive either.
> 
> At least for Solaris 2.5, they would not allow overallocation of swap...
> If you had pages that could be modified, you had to have enough swap
> storage to store a copy of those pages...  If you didn't, you'd get an
> out of memory error when trying to allocate the memory, such as forking,
> sbrk or mmap'ing...
> 
> FreeBSD has long allowed overallocation of swap because w/ early
> computers most people didn't have enough storage to handle it, and
> most memory won't be used...

There is a knob to disable overcommit in FreeBSD, see vm.overcommit
description in tuning(7). It is not very popular exactly because forking
large process requires reservation of twice memory comparing with the
real use.

I believe that the option is exercised by stress2, but I did not heard
about real-life usage, except my own long time ago.