Try the attached patch - it is against the 2.4.4 development tree, but it
should work in the release tree as well. I am still benchmarking and
scratching my head about this one, but it should help memory usage in
xfs. Is it possible to send the output of cat /proc/slabcache whilst
iozone is on its way to failing?
> I found your name and address on a thread about linux & xfs. I'm also
> seeing the "kernel: __alloc_pages: 0-order allocation
> failed." error on a 2.4.2-XFS system I just setup. We're comparing ext2,
> reiserfs, and xfs with iozone. The xfs performance was looking great (50%
> better on writes than ext2 or reiserfs) until the box fell over scrolling
> these errors. It hangs the box and only a hard reset will suffice to
> break out of it.
> I seem to be able to reproduce the error by running iozone this way:
> iozone -s 4000m
> That's a big iozone run, but the systems all have 2gig memory so I've got
> to go big to get out of linux cache and see meaningful numbers. Has there
> been any progress with this bug? If you'd like, I can get you a temporary
> acct. on the system where I'm testing.
> Galen Arnold, system engineer--systems group arnoldg@xxxxxxxxxxxxx
> National Center for Supercomputing Applications (217) 244-3473
> 152 Computer Applications Bldg., 605 E. Spfld. Ave., Champaign, IL 61820