Re: Process stuck in R state at extent = 2049

[Sakai Yasutoshi <sakai-yasutoshi@xxxxxxxxxxxxx>]

On 2004/12/07, at 9:13, Nathan Scott wrote:

> On Mon, Dec 06, 2004 at 11:24:34AM +0900, Sakai Yasutoshi wrote:
> > Hi.
> >
> > We are running linux 2.4.26 on IBM PPC405GP.
> >
> > Sometimes our process stuck in 'R' state when number of extent reaches
> > 2049.
> > And disk cache is freed at same time, so we have 40Mbyte of free 
> > memory,
> > but we get kernel message "kernel: __alloc_pages: 4-order allocation
> > failed (gfp=0xf0/0)".
> >
> > Process reverts to normal state after 5 seconds to 20 minutes once in 
> > a
> > while.
> >
> > is this XFS problem?
> Sort of; XFS has a tendency of asking for contiguous pages,
> and the VM has a tendency of not being able to satisfy such
> requests -- you might have more luck with a CVS 2.4 kernel,
> in particular the split-patches/cache_defs patch there adds
> a mechanism for 2.4 kernels to be able to do memory shaking
> which the XFS code can plug into.  2.6 provides a mechanism
> like that, without need for patching (fwiw).
It seems this patch fixes our problem.

Does XFS need more contiguous pages
especially when extents reachs 2049?

Thanks.

--
Yasutoshi Sakai