Re: __alloc_pages: 0-order allocation failed

[Andrea Arcangeli]

On Wed, Sep 26, 2001 at 11:15:09AM -0700, [e-mail] wrote:
> On Wed 26 Sep at 01:05:16 +0200 [e-mail] done said:
> > On Tue, Sep 25, 2001 at 06:25:10PM -0300, Marcelo Tosatti wrote:
> > > 
> > > Does vm-tweaks-1 fixes the current problem we're seeing? 
> > 
> > it seems no by reading the last email, however I'm not seeing any
> > problem, the DEBUG_GFP will tell us where the problem cames from,
> > pssobly it's a highmem thing since I never reproduced anything bad here.
> > But the point is that the above isn't going to be a right fix anyways.
> 
> vm-tweaks-1 fixes things for me.  I've got 512MB ram (kernel not
> configured for highmem) and 1 gig of swap.  The workload is heavy file
> i/o and has now been running almost 24 hours (about 2 billion I/Os or
> a few TB of data I think so far).  Previously all the memory was being
> consumed by cache, nothing swapped (as expected if the memory is cached
> buffer i/o right?) and I'd get the:

yes, unless the buffered I/O was identified as your very working set but
even in such case the 2.4.10 vm shouldn't swapout too early.

> 	__alloc_pages: 0-order allocation failed
> Now I continue to see the memory consumption / no swap, and no more
> error...iow the expected behaviour.

good. As far I can tell it is the check in swap_out that is making the
difference and fixing the oom problem, it was very intentional indeed.

> On an unrelated note if I want to backport the async I/O changes in 2.4.10,
> are there patches from you I should apply other than:
> 	2.4.10pre10aa1/40_blkdev-pagecache-17
> 	2.4.7pre8aa1/41_blkdev-pagecache-5_drop_get_bh_async-1

both patches are now included in mainline 2.4.10.

thanks,
Andrea