OOM-killer prevention for master kdeinit process
Lubos Lunak
l.lunak at suse.cz
Thu Aug 3 10:39:17 BST 2006
On Thursday 03 August 2006 10:37, Thomas Zander wrote:
> On Wednesday 2 August 2006 15:09, Lubos Lunak wrote:
> > short summary: Out Of Memory handling on Linux sucks, KDE can suffer
> > badly from it, kernel devels are unable to do anything about it, so
> > let's again work around things ourselves.
>
> Hi Lubos.
>
> Say; in what level does the existence of this kdeinit process depend on
> the lack of optimizations in binlib?
What do you mean with binlib?
> And do you have the idea that we might stop using it any time soon?
Yes, no (as in: I do, we might not). As far as speed is concerned we could
probably have dropped kdeinit by the time prelink became somewhat usable (on
the other hand it showed worthy even for things like the fontconfig shared
initialization hack). But when it comes to memory relocation processing
causes somewhere between 0,5-1M unshared memory per process (I don't remember
the numbers exactly, but it's closer to the 1). Prelink currently only
slightly reduces the number, -Bdirect and others won't at all, but kdeinit's
forking and copy-on-write takes care of most of it. If you run KDE with
KDE_IS_PRELINKED=1 the difference can be measured.
I don't know where to draw the line, but I myself would still like to keep
kdeinit for the time being, especially as I don't see any urging reasons to
drop it.
> For the ones that are not following; there actually are optimizations that
> the maintainer accepted:
> http://lwn.net/Articles/192624/
They're not _that_ big. 50% (as the subject of the linked mail says) out of a
lot is still quite a lot. They're kind of like improving bubblesort (even
though I'm still grateful even for them).
--
Lubos Lunak
KDE developer
---------------------------------------------------------------------
SuSE CR, s.r.o. e-mail: l.lunak at suse.cz , l.lunak at kde.org
Drahobejlova 27 tel: +420 2 9654 2373
190 00 Praha 9 fax: +420 2 9654 2374
Czech Republic http://www.suse.cz/
More information about the kde-core-devel
mailing list