On 953586655 seconds since the Beginning of the UNIX epoch
Chris G. Demetriou wrote:
>
>cgd@netbsd.org (Chris G. Demetriou) writes:
>> Michael Richardson <mcr@solidum.com> writes:
>> >   For the curious, linux-kernel is having a debate on overcommit.
>> > 
>> > http://www.uwsg.indiana.edu/hypermail/linux/kernel/0003.2/0003.html  
>> 
>> *chuckle*  Of course, we all individually know what the one true right
>> answer is...  (but damned if we could agree as a group... 8-)
>
>This statement, of course, has prompted several people (so far 8-) to
>ask me what The One True Right Answer (or at least, my answer) is.

I think that there would be some value to a semi-overcommit strategy
whereby pages that are COW are assigned a (dynamic) `likelihood'
of being copied, and that value is used to generate a likely maximum
of memory usage which will not be exceeded.  This would allow
typical system use to overcommit a little, but in general not run
out.  I would propose that this would replace a strict overcommit
strategy, and a non-overcommit would be provided.

 == Roland Dowdeswell                      http://www.Imrryr.ORG/~elric/  ==
 == The Unofficial NetBSD Web Pages        http://www.Imrryr.ORG/NetBSD/  ==
 == The NetBSD Project                            http://www.NetBSD.ORG/  ==