sbrk(2), OOM-killer and malloc() overcommit
Vadim Goncharov
vadimnuclight at tpu.ru
Fri Jan 4 10:32:41 PST 2008
04.01.08 @ 23:36 Tim Kientzle wrote:
> Vadim Goncharov wrote:
>> ... related "bug", in that of malloc overcommit,...
>
> malloc overcommit is not a bug; it's an important
> feature for many applications, for the same
> reasons that sparse files are an important feature.
> (Many applications can optimize performance by
> using an addressable region much larger than the
> actual data they need to store.)
What applications? How do they use this "feature"?
> If you really need a 4G block of memory, mmap()
> it to a file on disk.
That's not a solution for end-user. Why just not have space reserving, as
other operating systems can guarantee to their applications that stable
availability of memory?
--
WBR, Vadim Goncharov
More information about the freebsd-current
mailing list