Linux 2.4.4 folks

Linux 2.4.4 folks

[Peter Zaitsev]

Hello linux-kernel,

  I've trying to move some of my servers to 2.4.4 kernel from 2.2.x.
  Everything goes fine, notable perfomance increase occures, but the
  problem is I'm really often touch the following problem:

__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.
__alloc_pages: 1-order allocation failed.

This message may also show 1-order, 0-order, 3-order failures (only
one type at the time).  This problems also appeared then I tried to
use 2.4.1-2.4.3 kernels.

This sometimes leads to system hang, sometimes some processes gets
unkillable (even by kill -9) and in some cases I do not see any bad
results from this, but still this does not looks the right thing to
happen.

The problem is the systems this happens on are not short of memory.
Here is the free output for the system I had this happened this
morning:

rat:~ #  free
             total       used       free     shared    buffers     cached
Mem:       1028628    1025820       2808          0       9340     332412
-/+ buffers/cache:     684068     344560
Swap:      2097136          0    2097136


Does anyone has any ideas about this problem ?


-- 
Best regards,
 Peter                          mailto:pz@spylog.ru

Re: Linux 2.4.4 folks

[Mike Galbraith]

Hi,

On Sat, 19 May 2001, Peter Zaitsev wrote:

> Hello linux-kernel,
>
>   I've trying to move some of my servers to 2.4.4 kernel from 2.2.x.
>   Everything goes fine, notable perfomance increase occures, but the
>   problem is I'm really often touch the following problem:

<allocation failures snipped>

> The problem is the systems this happens on are not short of memory.
> Here is the free output for the system I had this happened this
> morning:

free doesn't show any info about allocator state at allocation failure
time.  If you were to modify mm/page_alloc.c:__alloc_pages() to do a
show_mem() along with the printk, you'd likely see that you're close
to being completely oom at that moment.

> Does anyone has any ideas about this problem ?

Posting /proc/slabinfo content might give the vm wizards an idea where
your memory is sitting.

If you see any indications of a sustained leak, I can provide you with
a 2.4.4 ikd patch (toolkit.. contains a memory leak detector) to try on
a non-mission-critical box.  It's large, 731003 uncompressed.

	-Mike

Re: Linux 2.4.4 folks

[Riley Williams]

Hi Peter.

 > I've trying to move some of my servers to 2.4.4 kernel from
 > 2.2.x. Everything goes fine, notable perfomance increase
 > occures, but the problem is I'm really often touch the following
 > problem:

 > __alloc_pages: 1-order allocation failed.
 > __alloc_pages: 1-order allocation failed.
 > __alloc_pages: 1-order allocation failed.
 > __alloc_pages: 1-order allocation failed.
 > __alloc_pages: 1-order allocation failed.

 > This message may also show 1-order, 0-order, 3-order failures
 > (only one type at the time).  This problems also appeared then I
 > tried to use 2.4.1-2.4.3 kernels.

 > This sometimes leads to system hang, sometimes some processes
 > gets unkillable (even by kill -9) and in some cases I do not see
 > any bad results from this, but still this does not looks the
 > right thing to happen.

 > The problem is the systems this happens on are not short of
 > memory. Here is the free output for the system I had this
 > happened this morning:

 > rat:~ #  free

 > Mem:       1028628    1025820       2808          0       9340     332412
 > -/+ buffers/cache:     684068     344560
 > Swap:      2097136          0    2097136

 > Does anyone has any ideas about this problem ?

I'm not up to date with the 2.4 series at the moment, but...

Looking at the figures you're showing, this looks like you have 1024M
of RAM. It used to be necessary to recompile the kernel if you had
more than (going from memory) 976M of RAM, where you had to change a
configuration option to select 2G of paging space instead of the
default 3G thereof, and this looks suspiciously like this problem to
me.

Can anybody confirm whether this limitation still applies?

Best wishes from Riley.