2002-01-18 15:09:59

by Anish Srivastava

[permalink] [raw]
Subject: kswapd kills linux box with kernel 2.4.17

Hi!

I am having a box with 8GB RAM and 8 CPU's.

I just changed updated the kernel to 2.4.17 and it booted up fine.....but as
soon as it started warming up
and using some memory.....kswapd came in and locked the CPU and the box
overall. The load went
upto about 29. It locked up the box and the CPU usage was at 99%

After sometime it did finish and I regained control on my system, but then
after some time the same problem

Can any of you help??

Thanks in anticipation.

Regards,
Anish Srivastava


2002-01-18 19:43:25

by Rik van Riel

[permalink] [raw]
Subject: Re: kswapd kills linux box with kernel 2.4.17

On Fri, 18 Jan 2002, Anish Srivastava wrote:

> I am having a box with 8GB RAM and 8 CPU's.

> Can any of you help??

There are two kernel patches which could help you, either
Andrea Arcangeli's VM patch (available from kernel.org)
or my -rmap VM patch (available from surriel.com/patches).

kind regards,

Rik
--
"Linux holds advantages over the single-vendor commercial OS"
-- Microsoft's "Competing with Linux" document

http://www.surriel.com/ http://distro.conectiva.com/

2002-01-19 13:53:14

by Anish Srivastava

[permalink] [raw]
Subject: Re: kswapd kills linux box with kernel 2.4.17

Hi! Rik,

Thanks for your help.
I have successfully patched the kernel with your patches
recompiled and installed it on my development system
the hardware of which is identical to my Production machine (e.g. 8CPU, 8GB
RAM)

I compiled the kernel with glibc-2.2.2-10 (redhat)
I did some performance testing on my box by doing a full oracle dump and
running some
big java jobs. Well Kswapd now seems to be behaving....(thankfully!!)

When I do a top it shows me
Mem: 8263740K av, 2967608K used, 5296132K free, 0K shrd, 6120K
buff
Swap: 2048248K av, 0K used, 2048248K free 2530948K
cached

Now the cached part never gets freed and just keeps piling up & so does the
used memory.
On my production box with (kernel 2.4.13), both cached & memory used
keet on increasing till it exhausts the entire physical RAM and the box
falls over.
It just doesn't swap.....(I thought with 8GB RAM I wouldnt need swap)

Anyways, then I updated the bdflush parameters....after that the memory does
get
reclaimed but only by a small percentage, so it hardly made any
difference..Memory still
keeps on piling up...forcing me to reboot the box everyday.

Do I need to make any changes to bdflush, freepages etc in /proc/sys/vm ??

Also, is it normal for linux to just keep on eating memory even though most
of the
processes are sleeping and not reclaim memory till the physical RAM is
exhausted

Once, again I thank you for your response and assistance. I really look
forward to
hearing from you again!!

Best regards,
Anish Srivastava

----- Original Message -----
From: "Rik van Riel" <[email protected]>
To: "Anish Srivastava" <[email protected]>
Cc: <[email protected]>
Sent: Saturday, January 19, 2002 1:12 AM
Subject: Re: kswapd kills linux box with kernel 2.4.17


> On Fri, 18 Jan 2002, Anish Srivastava wrote:
>
> > I am having a box with 8GB RAM and 8 CPU's.
>
> > Can any of you help??
>
> There are two kernel patches which could help you, either
> Andrea Arcangeli's VM patch (available from kernel.org)
> or my -rmap VM patch (available from surriel.com/patches).
>
> kind regards,
>
> Rik
> --
> "Linux holds advantages over the single-vendor commercial OS"
> -- Microsoft's "Competing with Linux" document
>
> http://www.surriel.com/ http://distro.conectiva.com/
>
>

2002-01-19 14:21:58

by Alan

[permalink] [raw]
Subject: Re: kswapd kills linux box with kernel 2.4.17

> Mem: 8263740K av, 2967608K used, 5296132K free, 0K shrd, 6120K
> buff
> Swap: 2048248K av, 0K used, 2048248K free 2530948K
> cached
>
> Now the cached part never gets freed and just keeps piling up & so does the
> used memory.

That is what I would expect up to a point. "Free" memory is wasted memory so
it is better to let all the free memory fill up with any bits of disk data
we have seen and might want again.

What actually matters (and sounds like 2.4.13 didnt do) is that at the point
you need memory for other things like applications and disk buffers that are
relevant to current usage the old stuff should rapidly get replaced by it.