kernel performance anomoly (help)

[Joe Elliott]

Hello Eddie,
	I see this anomoly with 2.4 kernels (except RedHat 7.2/3) releases)
It is quite well documented on the kernel dev groups. It effects the
kupdated daemon. This occurs when you have very high levels of IO on more
than 1 physical interface (ie eth/eth or eth/ide). kupdated can consume
100% cpu for 5/10 sec intervals and the machine crawls for a short period.
This repeats every minute or so while the high I/O persists.

You can easily see this with top. Ie run a big tcpdump capture with a 2.4
kernel (-s 1600 -w) and fill the wire with big packets. After the buffers
fill up you will see kupdated go nuts. Reading from the net and writing to
disk at high rates is 2.4's worst nitemare.

I have seen this with every 2.4 kernel to 2.4.19 (last one I tested)

Check the kernel.org newsgroup for details. Aparently 2.20 will be fixed,
but they said that about 2.18.
http://marc.theaimsgroup.com/?l=linux-kernel

I have heard from a few people that they have gone back to 2.2 kernels
for high I/O uses.

Then I discovered that RedHat 7.x fixes the problem! Who wants to do a diff
to find out what they did?

Joe.

-- 
                                          __o       _~o       __o
                                         `\<,      `\<,      `\<,
 ______________________________________(*)/_(*)__(*)/_(*)__(*)/_(*)________
 Im a 21st Century Digital Boy ... I aint got a life, but I got lotsa toys.
 *******************     Joe Elliott  joe at inetd.com    ********************
 Phone:(650)961-6631    Cell:(650)714-3932    Inetd.Com    http://inetd.com
 --------------------------------------------------------------------------

On Thu, 20 Jun 2002, root wrote:

> Date: Thu, 20 Jun 2002 05:42:13 -0400
> From: root <eric.freudenthal at nyu.edu>
> To: Eddie Kohler <kohler at icir.org>
> Cc: Brecht Vermeulen <brecht.vermeulen at rug.ac.be>,
>      Eric Freudenthal <eric.freudenthal at nyu.edu>,
>      Click Moil List <click at amsterdam.lcs.mit.edu>
> Subject: Re: kernel performance anomoly (help)
>
> Note that we see this slowdown for any 2.4 kernel we build, even without
> click.   Redhat's kernels have fine performance.
>
> Eric
>
>
> Eddie Kohler wrote:
>
> > > Well, I've seen also performance problems with 2.4.9, see
> > > http://www.pdos.lcs.mit.edu/click/ml/click/msg00977.html
> > > so I'm still sticking to 2.2.19.
> >
> > Does anyone have any idea what is causing this 2.4 slowdown?
> >
> > For instance, is _interrupting_ Click slower on 2.4 than on 2.2?
> > Or is it likely to be a polling-driver issue?
> >
> > Eddie
>