[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Swapper problems?

David Given wrote:
Hash: SHA1

When my Sheevaplug has been under load, I've been occasionally been
getting swapper page allocation failures (monster kernel log attached).

What's particularly worrying about this is that the system's got a
gigabyte of swap, and the log claims that it's all free. So why should
it be behaving as if it's running out of memory? Given that I want to
run Java on this machine, this could be an issue!

This problem is, in fact, strikingly similar to recurrent issue on
Debian on the NSLU2, where when the system was under heavy load the OOM
killer would erroneously start nuking processes at random, regardless of
the amount of available swap. Are there any known problems with the VM
on ARM?

I'm using:

Linux chur 2.6.30-rc8-kirkwood #1 Wed Jun 3 22:03:05 UTC 2009 armv5tel

The swap is a file on a JFS partition on a RAID0 volume on USB, if that
helps. I'm about to retry with 2.6.31-rc7 and see if that helps.
Try running the following as root:
sysctl -w overcommit_memory=2
sysctl -w overcommit_ratio=95

Ref: http://www.redhat.com/magazine/001nov04/features/vm/

That should set a sane memory allocation policy and prevent the OOM killer. The downside is that a program will probably fail because it can't allocate enough memory.


Reply to: