Re: 2.6.20-rc5: cp 18gb 18gb.2 = OOM killer, reproducible just like 2.16.19.2

[Bill Cizek <cizek@rcn.com>]

Justin Piszcz wrote:
> On Mon, 22 Jan 2007, Andrew Morton wrote:
>   
>>> On Sun, 21 Jan 2007 14:27:34 -0500 (EST) Justin Piszcz <jpiszcz@lucidpixels.com> wrote:
>>> Why does copying an 18GB on a 74GB raptor raid1 cause the kernel to invoke 
>>> the OOM killer and kill all of my processes?
>>>       
> Running with PREEMPT OFF lets me copy the file!!  The machine LAGS 
> occasionally every 5-30-60 seconds or so VERY BADLY, talking 5-10 seconds 
> of lag, but hey, it does not crash!! I will boot the older kernel with 
> preempt on and see if I can get you that information you requested.
>   
Justin,

According to your kernel_ring_buffer.txt (attached to another email), 
you are using "anticipatory" as your io scheduler:
   289  Jan 24 18:35:25 p34 kernel: [    0.142130] io scheduler noop 
registered
   290  Jan 24 18:35:25 p34 kernel: [    0.142194] io scheduler 
anticipatory registered (default)

I had a problem with this scheduler where my system would occasionally 
lockup during heavy I/O.  Sometimes it would fix itself, sometimes I had 
to reboot.  I changed to the "CFQ" io scheduler and my system has worked 
fine since then.

CFQ has to be built into the kernel (under BlockLayer/IOSchedulers).  It 
can be selected as default or you can set it during runtime:

echo cfq > /sys/block/<disk>/queue/scheduler
...

Hope this helps,
Bill