On Wed, 16 Jan 2002 09:33:42 +0000 Sean Neakums <Sean Neakums
<sneakums@xxxxxxxx>> wrote:
> begin Dan Hollis quotation:
> > Its too bad theres no signal mechanism for userspace for 'memory
> >low', eg SIGLOWMEM so userspace apps could voluntarily release
> >resources to ease memory pressure.
> There was a big discussion on LKML about that a few months back. I
> forget what the outcome was.
Implementation on these things can be crucial.
A few years ago I was having trouble with Oracle on an ICL Unix box
(though I think the vendor was immateriel). periodically when it tried
to get extra resources it would go into a tight loop, grabbing all
available CPU. We eventually concluded it was calling malloc, not
getting as much as it wanted, calling malloc........ . I'm sure you get
the picture.
--
Keith Matthews
Frequentous Consultants - Linux Services,
Oracle development & database administration
|